{"id":"974373af-2d44-40f3-a212-8f3ced1493cd","arxiv_id":"1908.06505","paper_version":3,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Null-space projection of the transmit precoder suppresses self-interference in mmWave full-duplex, and hybrid-aware designs bring the sum rate near the ideal full-duplex bound.","lead":"This paper presents beamforming cancellation designs that let a millimeter-wave device transmit and receive in the same band by steering its beams to cancel self-interference. Simulations show the sum spectral efficiency approaches the self-interference-free ideal, under two hybrid beamforming scenarios.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Case B's null-space projection cannot support the promised Ns streams when NRF<2Ns; the effective SI channel has too few null dimensions, so the practical design cannot approach SI-free FD sum rate even with perfect CSI.","rationale":"The reader's weakest assumption was perfect CSI of Hii, and that is a legitimate practical concern. However, the more load-bearing issue is internal to the Case B construction: the null space of the effective SI channel is too small to hold Ns independent precoder columns whenever NRF < 2Ns. The paper's own equations imply that F_BB must be a solution of C_eff F_BB = 0, with C_eff of size Ns x NRF. For a generic channel, the nullity is NRF-Ns, which is less than Ns in the stated regime. Therefore the projected digital precoder has rank deficiency and cannot support the assumed Ns streams to (j). This directly invalidates the central claim that the practical Case B design approaches SI-free FD spectral efficiency, even under the paper's ideal-CSI assumptions. The reader noticed the boundary case NRF=Ns where the null space is empty, but the problem is broader: it affects the entire stated range Ns <= NRF < 2Ns. Because the design as written cannot meet its claimed performance in the regime it advertises, the appropriate verdict is REJECT, unless the authors substantially revise the parameter regime or the stream-count claim. The dimension-count argument is easy to verify and does not depend on simulation details, code availability, or CSI robustness.","tokens_in":9001,"tokens_out":14790,"duration_ms":151096,"concrete_test":"Run the Case B algorithm exactly as described (OMP decomposition, then F_BB = Q XBB) for Ns=3 with NRF=4 and NRF=5, using the Section IV channel model; compute rank(F_BB) and the singular values of the effective desired channel H_ij F_RF F_BB. If rank(F_BB) < Ns (e.g., 1 for NRF=4) and fewer than Ns nonzero singular values are obtained, the (i)->(j) link cannot deliver Ns streams and the sum spectral efficiency cannot approach the SI-free FD curve. As an analytic cross-check, verify that dim null(W_BB^H W_RF^H Hii F_RF) equals NRF-Ns for generic random channels and compare this dimension with the number of columns of F_BB that must lie in the null space.","verdict_should_be":"REJECT","load_bearing_attack":"In Case B (Section III, 2Ns > NRF >= Ns), the design requires the NRF x Ns digital precoder F_BB to satisfy C_eff F_BB = 0, where C_eff = W_BB^H W_RF^H Hii F_RF (Eq. 18) is Ns x NRF. For generic full-rank mmWave channels and the previously chosen eigen-combiner and OMP analog precoder, rank(C_eff)=Ns, so dim null(C_eff)=NRF-Ns. Under the stated assumption 2Ns > NRF, this nullity is strictly less than Ns. But F_BB = Q XBB (Eqs. 19-20) must place its Ns columns inside that null space; hence rank(F_BB) <= NRF-Ns < Ns. Example: Ns=3, NRF=4 gives a one-dimensional null space and a rank-1 transmit precoder, so the (i)->(j) link can carry at most one spatial stream. Consequently the link's spectral efficiency at high SNR is reduced to roughly (NRF-Ns)log(SNR) rather than Ns log(SNR), and the sum spectral efficiency cannot approach that of an SI-free FD system as claimed. This is independent of CSI error; it is a dimensional impossibility. The paper never checks dim null(C_eff) against Ns, and the Fig. 4/5 discussion of 'offering more dimensions to F_BB' misses that for NRF<2Ns those dimensions are always insufficient for Ns streams. Only at NRF=2Ns does the nullity equal Ns, which is the Case A boundary, not Case B.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper proposes beamforming cancellation (BFC) designs for in-band full-duplex millimeter-wave communication with hybrid beamforming at the full-duplex node (i). The node uses separate transmit and receive arrays and seeks to null self-interference by projecting its transmit precoder onto the null space of the effective SI channel observed through its receive combiner. Case A assumes NRF=2Ns RF chains with infinite-resolution phase shifters, enabling perfect hybrid decomposition of the projected fully-digital precoder. Case B treats the more practical regime 2Ns > NRF >= Ns with finite-resolution phase shifters, using orthogonal matching pursuit to obtain the analog precoder and then projecting only the digital precoder into the null space of the resulting effective SI channel. The paper claims that in both cases the sum spectral efficiency approaches that of an SI-free full-duplex system, and it presents simulations for Ns=3 with different array sizes and RF chain counts.","tokens_in":9185,"tokens_out":8151,"duration_ms":76695,"significance":"The conceptual approach of nulling the effective SI channel rather than the full SI channel is useful, and the Case A design is mathematically sound: with NRF=2Ns, the null space of the Ns x Nt effective channel has dimension at least Ns for Nt>Ns, so the projected precoder can preserve Ns streams. The paper also clearly demonstrates that eigen-beamforming alone is insufficient for mmWave FD, and it offers a non-iterative design that does not require joint optimization across nodes. These are strengths. However, the central claim about Case B fails under the stated parameter range because the null space of the effective SI channel is too small to carry Ns streams; this is a dimensional impossibility, not a matter of optimization quality, and it undermines the paper's main practical contribution.","major_comments":[{"comment":"The dimension of the null space of the effective SI channel is insufficient to support Ns streams when NRF < 2Ns. Since C_eff in (18) is Ns x NRF and generically has rank Ns, its null space has dimension NRF - Ns < Ns under the stated condition 2Ns > NRF. The projected digital precoder F_BB = Q XBB in (20) therefore has rank at most NRF - Ns, so the link from (i) to (j) can carry at most NRF - Ns streams rather than Ns. For the example Ns=3, NRF=4 used in the simulations, this limits the transmit link to one stream, so the sum spectral efficiency cannot approach that of an SI-free FD system at high SNR; the claim in the abstract and in Section IV that both designs approach ideal FD performance is not supported.","section":"Section III, Case B, Eqs. (18)-(20)"},{"comment":"The discussion of 'offering more dimensions to the digital precoder F_BB' as NRF increases overlooks that for every NRF < 2Ns, the nullity remains less than Ns, so the transmit link is rank-deficient regardless of the hybrid approximation quality. The simulations in Figs. 4-5 should report the number of spatial streams actually delivered after the projection; otherwise the spectral efficiency results are difficult to interpret and the comparison with the NRF=6 eigen-precoder baseline is misleading.","section":"Section IV, Results of Case B"},{"comment":"The conclusion states that the designs 'achieve significant spectral efficiency gains approaching that of ideal FD operation.' This is accurate for Case A (NRF=2Ns) and for Case B only at the boundary NRF=2Ns, which is excluded from the stated Case B range. As written, the conclusion overstates the practical Case B, which suffers an unavoidable stream deficit for NRF<2Ns.","section":"Section V, Conclusion"}],"minor_comments":[{"comment":"The motivation for considering the effective SI channel could be complemented by an explicit statement that the null space dimension of the effective channel must be at least Ns for the projection in (16) or (20) to preserve the number of streams; this would have highlighted the Case B limitation.","section":"Section III, after Eq. (14)"},{"comment":"The paper assumes perfect CSI of the SI channel; a sentence acknowledging the sensitivity to CSI error and possible extensions would strengthen the practical claims.","section":"Section III, Assumptions"},{"comment":"The captions of Figs. 4 and 5 should specify the exact NRF values and the corresponding supported stream counts after projection, since the text leaves this ambiguous.","section":"Figs. 4 and 5"},{"comment":"The notation F(i) is used both for the fully-digital precoder in (16) and for the effective hybrid precoder in (21); consider distinguishing these two quantities.","section":"Notation"},{"comment":"The expression for the phase shift, phi(theta) = 2*pi/lambda_c * lambda_c/2 * cos(theta), simplifies to pi*cos(theta); the simplification would be clearer.","section":"Section II, array response"}],"recommendation":"major_revision","confidential_remarks":"The dimensional flaw in Case B is substantial and, if left unaddressed, would make the paper unsuitable for publication as is. However, the underlying projection idea is salvageable. A major revision that restricts Case B to NRF >= 2Ns or thoroughly characterizes the stream reduction, and revises the abstract and conclusion accordingly, could make the work acceptable. The editor may wish to seek an additional reviewer with expertise in MIMO rank analysis to verify the stream-count argument."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear [Colleague],\n\nQuick take: the paper's core idea—project the transmit precoder into the null space of the effective SI channel—is a legitimate zero-forcing trick, and it is worked out cleanly for the NRF=2Ns case. But the second case, the one the authors present as practical (2Ns > NRF >= Ns), is broken by a simple dimension count.\n\nWhat's new and good: the problem is real. Prior BFC work was iterative, fully-digital, and ignored hybrid constraints. This paper gives a non-iterative construction that respects the constant-amplitude RF constraint, uses the eigen-combiner to define an effective SI channel, and projects the precoder into its null space. Under perfect CSI, the SI term is zero by construction in both cases. Case A (NRF=2Ns, infinite-phase resolution) is a valid instantiation: with NRF=2Ns the null space of the Ns x NRF effective channel has dimension Ns, so F_BB can carry Ns independent streams. That part is correct and worth knowing.\n\nThe soft spot is not CSI robustness; it's dimensional. In Case B, after fixing W_BB, W_RF, and F_RF, the effective SI channel C_eff = W_BB^H W_RF^H Hii F_RF is Ns x NRF. For generic channels it has rank Ns, so its null space dimension is NRF - Ns. For any NRF < 2Ns, that's strictly less than Ns. The design sets F_BB = Q X_BB, so every column of F_BB lies in that null space and rank(F_BB) <= NRF - Ns. The link to (j) therefore carries at most NRF - Ns streams, not Ns. With Ns=3, NRF=4, you get one stream. The sum spectral efficiency at high SNR is about (Ns + NRF - Ns) log SNR = NRF log SNR, not 2 Ns log SNR. That is not approaching the SI-free FD sum rate. The paper never checks nullity against Ns, and the claim that more RF chains \"offer more dimensions to F_BB\" only makes sense up to NRF=2Ns; at NRF=2Ns you are in Case A's boundary, not Case B.\n\nAlso minor: perfect CSI for Hii at 120 dB SI is assumed with no error analysis, and no code, data, or error bars are given. Those would be fixable if the dimension issue were addressed.\n\nBottom line: Case A is a sound, publishable observation. Case B as written is not. A serious referee should send it back for major revision, or the authors should reframe Case B as supporting NRF - Ns streams, which is a much weaker result. I would send it to review because the problem and the Case A construction deserve referee time, but the revision has to confront the dimension count.","headline":"The null-space projection idea is sound for NRF=2Ns, but the more practical Case B cannot support Ns transmit streams for NRF<2Ns, so its headline claim fails on a dimension count.","tokens_in":9850,"tokens_out":6614,"would_cite":false,"duration_ms":63595,"reading_group":"no","serious_thinker":"no","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Projecting the transmit beam into the null space of the effective self-interference channel cancels a mmWave full-duplex node's own interference, bringing its sum spectral efficiency close to ideal.","keywords":["millimeter-wave","full-duplex","beamforming cancellation","self-interference","hybrid beamforming","null-space projection","spectral efficiency","MIMO"],"falsifier":"Run either proposed design in simulation or over the air using an SI channel estimate corrupted by a known error (for example, Gaussian noise added to $\\mathbf{H}_{ii}$); if the residual self-interference power after projection exceeds the noise floor for error levels typical of mmWave channel estimation, the claim of complete SI elimination is falsified under imperfect CSI, which is the practical condition.","tokens_in":8649,"feed_emoji":"📡","tokens_out":10913,"duration_ms":92915,"temperature":0.7,"pith_summary":"This paper sets out to show that millimeter-wave full-duplex communication can be made practical by cancelling self-interference purely through beam steering, without the iterative cancellation methods used below 6 GHz. The authors propose two beamforming designs for a fully-connected hybrid array: one for ideal hardware with twice as many RF chains as streams and infinite-resolution phase shifters, and one for realistic finite-resolution phase shifters with fewer RF chains. Both designs fix the receive combiner and then project the transmit beam into the null space of the effective self-interference channel, the interference as seen after combining, which makes the received self-interference exactly zero. Simulation results show the sum spectral efficiency approaching that of an ideal full-duplex system with no self-interference. The significance is that in-band simultaneous transmit and receive at mmWave would double spectral efficiency and reduce latency, and existing sub-6 GHz self-interference cancellation methods do not translate to mmWave.","feed_headline":"MmWave full-duplex approaches ideal rates with null-space beam steering","feed_subtitle":"By projecting the transmit beam into the interference channel's null space, one node can transmit and receive in-band.","key_machinery":"The load-bearing object is the effective self-interference channel, $\\mathbf{W}_{BB}^{(i)H}\\mathbf{W}_{RF}^{(i)H}\\mathbf{H}_{ii}\\mathbf{F}_{RF}^{(i)}$ in Case B (reduced to $\\mathbf{W}^{(i)H}\\mathbf{H}_{ii}$ in Case A), which is the portion of the SI channel that survives the receive combiner. The mechanism is the orthogonal projection matrix $P = B(B^H B)^{-1}B^H$, where $B$ is a basis of that channel's null space; applying $P$ to the transmit precoder forces the SI term in the received signal to zero while keeping the beam as close as possible to the eigen-precoder for the desired link. This works because although the raw SI channel is full rank and has no null space, the effective channel seen after combining has a large null space when the number of streams is well below the number of antennas.","core_discovery":"The paper's central claim is that the received self-interference at a full-duplex node can be completely eliminated by design, not by estimation and subtraction. Fixing the receive combiner to the eigen-combiner for the link from $k$, the authors define the effective self-interference channel as the SI channel seen through that combiner, and require the transmit precoder to lie in its null space so that $\\mathbf{W}^{(i)H}\\mathbf{H}_{ii}\\mathbf{F}^{(i)} = \\mathbf{0}$. In Case A, with $N_{\\mathrm{RF}} = 2N_s$ and infinite-resolution phase shifters, the fully-digital eigen-precoder is projected onto the null space and then decomposed perfectly into hybrid form. In Case B, with $2N_s > N_{\\mathrm{RF}} \\geq N_s$ and finite resolution, the projection is applied to the digital precoder after an OMP-based hybrid decomposition, so the constraint is enforced inside the hybrid architecture. The consequence is that the link from $k$ to $i$ achieves its half-duplex spectral efficiency while $i$ simultaneously serves $j$, and the sum spectral efficiency hugs the ideal full-duplex curve; the only cost is a slight deviation of the transmit beam from the unconstrained eigen-precoder. The paper also shows that eigen-beamforming alone fails completely, because even a tiny fraction of the 120 dB self-interference dominates the desired signal.","pith_inferences":["A direct extension would test robustness to imperfect SI channel knowledge: because the projection is exact only with perfect CSI, practical deployment would need either estimation-error-aware projection or a residual analog/digital cancellation stage.","The same null-space idea should carry over to a full-duplex node serving multiple users or streams, with each stream projected into the effective SI null space; the cost in desired-link rate will grow as the number of streams approaches the number of antennas.","Fixing the receive combiner to the eigen-combiner, as the paper does, is a choice; jointly optimizing the combiner and the projected precoder could yield a higher sum rate than this design, since the combiner determines the effective SI channel and hence the available null space.","Because the cancellation is purely spatial, it is complementary to existing self-interference cancellation techniques; a practical mmWave full-duplex system would likely combine the two to relax the perfect-CSI assumption."],"forward_implications":["A single mmWave node can serve a downlink user and receive from an uplink user in the same band at the same time, with a sum spectral efficiency close to the theoretical self-interference-free full-duplex bound.","Eigen-beamforming alone cannot support mmWave full-duplex: the 120 dB self-interference overwhelms the desired receive signal, so a deliberate null-space projection is essential.","Increasing the number of RF chains in the practical Case B design mainly adds dimensions for null-space projection, which improves the transmit link more than it improves hybrid approximation accuracy.","Larger antenna arrays enlarge the null space of the effective SI channel, so the transmit beam can stay closer to the unconstrained eigen-precoder while still cancelling self-interference exactly.","The designs are non-iterative and do not require joint optimization across the transmitter and receiver, unlike earlier mmWave full-duplex beamforming-cancellation approaches."],"supporting_citations":[{"why":"Supplies the Rician LOS/NLOS model for the mmWave SI channel and the prior iterative BFC designs that this paper's closed-form projection replaces.","marker":"[7]"},{"why":"Shows that with 2Ns RF chains and infinite-resolution phase shifters any fully-digital beamformer can be decomposed perfectly into hybrid form, which Case A relies on.","marker":"[11]"},{"why":"Provides the OMP-based hybrid decomposition with an analog codebook used in Case B to approximate the fully-digital precoder before null-space projection.","marker":"[13]"},{"why":"Gives the spherical-wave LOS entry model for the near-field SI channel used in equation (7).","marker":"[12]"},{"why":"Provides the fully-connected hybrid MIMO architecture and the extended Saleh-Valenzuela mmWave channel model used for the desired links and NLOS SI.","marker":"[4]"},{"why":"Establishes the mmWave MIMO background that motivates dense arrays and hybrid beamforming, framing the constraints the designs must satisfy.","marker":"[2]"}],"fun_headline_variants":["Null-space beams unlock full-duplex mmWave","Designer zeros erase self-interference in mmWave FD","Beamforming cancellation hits ideal full-duplex rates","Steer transmit beam into null space for FD mmWave","MmWave FD: zero self-interference by design"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The designs assume the full-duplex node knows its own self-interference channel exactly; if the channel estimate is even slightly wrong, the projected beam will not lie in the true null space, and the 120 dB self-interference will overwhelm the desired receive signal.","fun_headline_variants_meta":{"raw":{"variants":["Null-space beams unlock full-duplex mmWave","Designer zeros erase self-interference in mmWave FD","Beamforming cancellation hits ideal full-duplex rates","Steer transmit beam into null space for FD mmWave","MmWave FD: zero self-interference by design"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000198,"raw_usage":{"total_tokens":1392,"prompt_tokens":995,"completion_tokens":397,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":611,"completion_tokens_details":{"reasoning_tokens":320}},"tokens_in":611,"tokens_out":397,"duration_ms":4484,"temperature":1.0,"reasoning_tokens":320,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T12:44:02.027700+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run either proposed design in simulation or over the air using an SI channel estimate corrupted by a known error (for example, Gaussian noise added to $\\mathbf{H}_{ii}$); if the residual self-interference power after projection exceeds the noise floor for error levels typical of mmWave channel estimation, the claim of complete SI elimination is falsified under imperfect CSI, which is the practical condition.","supporting_citations":[{"cited_title":"Hybrid beamforming design for full-duplex millimeter wave communication,","cited_arxiv_id":null,"evidence_quote":"Supplies the Rician LOS/NLOS model for the mmWave SI channel and the prior iterative BFC designs that this paper's closed-form projection replaces."},{"cited_title":"Hybrid digital and analog beamforming design for large-scale antenna arrays,","cited_arxiv_id":null,"evidence_quote":"Shows that with 2Ns RF chains and infinite-resolution phase shifters any fully-digital beamformer can be decomposed perfectly into hybrid form, which Case A relies on."},{"cited_title":"Spatially sparse precoding in millimeter wave MIMO systems,","cited_arxiv_id":null,"evidence_quote":"Provides the OMP-based hybrid decomposition with an analog codebook used in Case B to approximate the fully-digital precoder before null-space projection."},{"cited_title":"Spherical-wave model for short-range MIMO,","cited_arxiv_id":null,"evidence_quote":"Gives the spherical-wave LOS entry model for the near-field SI channel used in equation (7)."},{"cited_title":"An overview of signal processing techniques for millimeter wave MIMO systems,","cited_arxiv_id":null,"evidence_quote":"Establishes the mmWave MIMO background that motivates dense arrays and hybrid beamforming, framing the constraints the designs must satisfy."}],"review_version":1}