{"id":"ef460f60-62af-407d-bb15-db62f686157d","arxiv_id":"1908.03359","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"Coordinated constructive-interference hybrid precoding with MILP-based RF chain assignment reduces total transmit power relative to zero-forcing and uncoordinated CI precoding in simulated heterogeneous massive MIMO networks.","lead":"This paper proposes a coordinated hybrid precoding scheme for heterogeneous cellular networks in which base stations deliberately exploit interference between users instead of canceling it, aiming to cut transmit power while meeting quality-of-service targets.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Reported power savings may be an artifact of using a single-cell CI SER-vs-TNR curve to evaluate the ZF baselines.","rationale":"The reader's weakest assumption identifies exactly the same load-bearing concern: the empirical SER mapping from [28] is applied to both the proposed CI schemes and the ZF baselines without justification. This is a genuine methodological risk because the comparison in Fig. 3 is the sole support for the paper's central claim of superior performance. The paper provides no theoretical performance bound and no direct symbol-level simulation, so the reported power savings depend entirely on the validity of that mapping. The concern is concrete and testable: a direct Monte Carlo SER evaluation would settle whether the ordering and magnitude of the gains are real. I agree with the reader's conditional verdict, as the mathematical formulation and algorithm appear standard and the issue is one of verification rather than a fundamental flaw in the approach. No adjustment to the verdict is warranted; the condition (direct SER simulation or a theoretical justification of the mapping) should be part of the final acceptance criteria.","tokens_in":9256,"tokens_out":3746,"duration_ms":39469,"concrete_test":"Run direct Monte Carlo symbol-level simulations for the same setup: for each channel realization, compute the hybrid precoders from the proposed three-stage algorithm and from the ZF baselines, transmit the resulting QPSK symbols, add AWGN with the same noise power, perform ML detection, and count symbol errors across at least 10,000 realizations. Plot the measured SER versus total transmit power for Coord. CI (continuous), Coord. ZF (continuous), and Uncoord. CI (continuous). If the ordering and the gap at SER=10^-4 match Fig. 3, the indirect mapping is validated; if not, the claim of superior performance is not established.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The headline performance claim in Fig. 3 is not supported by direct symbol-level simulation. As stated in footnote 4 of Section IV, the SER values for both the proposed CI precoders and the ZF baselines are computed by mapping the simulated TNR/SNR to SER using the empirical curve in Fig. 10 of [28], which was obtained for the authors' earlier single-cell CI hybrid precoding. This transfer assumes (i) that the SER-vs-TNR relationship for CI is invariant to the coordinated multi-BS setting, the continuous/codebook analog precoding, and the suboptimal three-stage solution, and (ii) that the same empirical curve correctly describes SER for zero-forcing precoding, whose error probability follows the standard post-detection SNR law (e.g., approximately 2Q(sqrt(SNR)) for QPSK). Neither assumption is justified in the paper. If Fig. 10 of [28] was calibrated for CI only, the ZF curves in Fig. 3 are shifted by an unknown amount, and the reported power savings at SER=10^-4 could be an artifact of the evaluation procedure. The missing simulation parameters (codebook size, epsilon in (4), per-BS user counts) further prevent independent reproduction.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a coordinated hybrid precoding scheme for a downlink multiuser massive MIMO heterogeneous network. The base stations, connected to a central controller, jointly exploit constructive interference rather than suppress it, with the goal of minimizing total transmit power while satisfying per-user quality-of-service thresholds. The problem is formulated as a nonconvex optimization and solved via a three-stage heuristic: an MILP-based RF-chain/user assignment, analog precoder design (either continuous or codebook-based), and a convex digital precoding stage. Simulation results in Figure 3 are used to claim that the coordinated CI-based hybrid precoders require substantially lower total transmit power than coordinated zero-forcing and uncoordinated CI baselines at the same symbol error rate. A backhaul-overhead analysis in Figure 4 shows that the CI method exchanges fewer coefficients and symbols than ZF.","tokens_in":9533,"tokens_out":3576,"duration_ms":39891,"significance":"The idea of coordinating multiple base stations to exploit inter-cell interference in hybrid precoding is timely and, if valid, would represent a useful step toward energy-efficient dense networks. The problem reformulation in equations (2)-(3) is clean, and formulating the RF-chain assignment as an MILP for both continuous and codebook analog precoding is a reasonable design contribution. However, the central performance claim currently rests on an indirect evaluation method that maps empirical SNR/TNR-to-SER curves from a prior single-cell study to the coordinated multi-BS setting and to the ZF baselines. Because this mapping is not justified for either the proposed method or the baselines, the numerical evidence is not convincing. The paper would be significantly improved by direct symbol-level simulations or a validated analytical error-probability model for each compared scheme.","major_comments":[{"comment":"The SER values for all plotted schemes, including the ZF baselines, are computed by mapping the simulated TNR or SNR to SER using the empirical curve from Fig. 10 of [28], which was obtained for the authors' earlier single-cell CI hybrid precoding. This is not a valid comparison procedure. The SER-vs-SNR relationship for zero-forcing precoding is fundamentally different from the SER-vs-TNR relationship for CI-based precoding (for QPSK, ZF roughly follows 2Q(sqrt(SNR)) while CI follows a different law tied to the threshold margin). Applying the CI-derived curve to ZF results in an unknown shift of the ZF curves in Fig. 3, and the reported power savings at SER=10^-4 may be an artifact of this shift rather than a genuine advantage of the proposed scheme. The authors must replace this indirect mapping with direct symbol-level simulations for every scheme, or at minimum use the theoretical Q-function relationship for ZF and a separately validated CI SER curve for the coordinated setting.","section":"Section IV, Fig. 3 and footnote 4"},{"comment":"Even for the proposed coordinated CI schemes themselves, the empirical SER-vs-TNR curve from [28] was obtained for a single-cell, non-coordinated system with a specific antenna configuration and analog precoding method. The coordinated multi-BS setup in this paper has different per-BS antenna counts (N_macro=64, N_pico=32), a different number of users (K=64), and a different suboptimal three-stage solution, all of which can alter the actual SER-vs-TNR relationship. No argument or measurement is given that the [28] curve transfers to this new setting. Consequently, the quantitative gain of the proposed scheme over the baselines is not established by the current evidence.","section":"Section IV, Fig. 3 and footnote 4"},{"comment":"The paper provides no theoretical performance bound or optimality gap analysis for the proposed three-stage decomposition. Since the original problem (2) is nonconvex and the solution is suboptimal, the claim that the scheme achieves 'superior performance' depends entirely on the numerical results. Given that the numerical results are obtained through the problematic empirical mapping described above, the central claim is currently unsupported. The authors should either prove a performance guarantee (e.g., a bound on the transmit power relative to the optimal solution) or supply direct simulation results that verify the SER curves in Fig. 3.","section":"Section III and IV"}],"minor_comments":[{"comment":"The simulation section omits several parameters needed for reproducibility: the size of the analog codebook and its design, the value of the fairness scaling factor epsilon in (4) and (5), the per-BS user distribution or user-association rule for the uncoordinated scheme, and the specific TNR values swept for the CI methods. Please provide these details.","section":"Section IV"},{"comment":"The channel vector h_gk is used with a transpose (h^T) in the received signal model, but for complex baseband channels the Hermitian transpose (h^H) is the standard form. Please clarify the notation or correct it to avoid confusion.","section":"Equation (1)"},{"comment":"The figure caption for Fig. 3 does not mention that the SER values are derived from the empirical mapping in footnote 4 rather than from direct symbol error counting. The axis label 'SER' should be qualified (e.g., 'estimated SER') to alert the reader to the indirect evaluation.","section":"Section IV"},{"comment":"The results are averaged over 10,000 Monte Carlo runs, but no confidence intervals or error bars are shown. Given the indirect SER estimation, it would be useful to indicate the variance of the estimated SER across runs.","section":"Section IV"},{"comment":"The paper assumes a fully-connected hybrid architecture with analog precoding coefficients of identical magnitude a, but the value of a is not specified. Please state whether a is set to unit magnitude or some other value, and confirm that the power constraint (2c) accounts for this normalization.","section":"Section II"}],"recommendation":"major_revision","confidential_remarks":"The main concern is the evaluation methodology in Fig. 3: both the proposed CI methods and the ZF baselines are evaluated using an empirical SER-vs-TNR/SNR curve from the authors' own prior work. This creates a self-referential loop that is difficult to validate and is likely to be challenged by reviewers and readers. The paper would be much stronger if the authors reran the comparison with direct symbol-level simulations, even at reduced scale, or used a validated theoretical SER formula for ZF. As a letter, space is limited, but the performance claim is the core of the paper and must be supported by credible evidence. I recommend sending the manuscript back for major revision with a request to replace or supplement the indirect SER mapping."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe useful part of this paper is the framework: coordinated constructive-interference hybrid precoding across heterogeneous BSs, with a MILP for RF chain and codebook assignment. The problem reformulation in (2)-(3) is clean, the three-stage decomposition is sensible, and the digital stage is convex. That is worth taking seriously.\n\nThe soft spot is the evaluation. Figure 3, which carries the entire performance claim, computes SER for both the proposed CI schemes and the ZF baselines by mapping simulated TNR/SNR to SER using the empirical curve from Fig. 10 of the authors' earlier single-cell CI paper [28]. Footnote 4 says this explicitly. The problem is obvious: for ZF, the SER as a function of SNR is a standard Q-function curve; using a CI-calibrated empirical mapping may shift the ZF curves by an unknown amount. Unless the authors show that mapping is valid for ZF and for coordinated transmission, the reported transmit power savings at SER=10^-4 are not established. Missing simulation parameters (codebook size, the epsilon in (4)/(5), per-BS user counts) make independent reproduction difficult.\n\nI do not think the math is broken, and the MILP assignment formulation is a legitimate new combination of known ingredients. The central claim, however, currently rests on an untested transfer. The fix is straightforward: run direct symbol-level simulations for at least the ZF baselines, or provide a theoretical justification for the mapping. That is a moderate revision, not a rewrite.\n\nWho is this for? Researchers working on CI-based precoding or hybrid precoding in coordinated networks. They should treat Figure 3 as unverified until direct simulation appears. As it stands, I would take the formulation, not the numbers.\n\nRecommendation: this deserves a serious referee, because the framework is solid and the flaw is in the evaluation rather than the ideas. The referee should ask for direct SER simulation before publication.","headline":"Clean extension of CI hybrid precoding to HetNets, but the headline power-savings claim rests on an unvalidated reuse of a single-cell CI SER curve for the ZF baselines.","tokens_in":10050,"tokens_out":2337,"would_cite":false,"duration_ms":23941,"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":"Coordinated hybrid precoding exploits interference to cut transmit power.","keywords":["constructive interference","hybrid precoding","heterogeneous networks","coordinated multipoint","massive MIMO","power minimization","mixed-integer linear programming","energy efficiency"],"falsifier":"Perform end-to-end Monte Carlo simulation in the same three-base-station heterogeneous network (macro with 64 antennas, two picos with 32 antennas, 64 QPSK users), transmitting actual symbols through the proposed coordinated CI hybrid precoder and the coordinated zero-forcing baseline, detecting with a maximum-likelihood rule, and plotting symbol error rate against transmit power; if zero-forcing achieves the same symbol error rate at lower or equal power than the CI scheme, or if the CI scheme's measured symbol error rate does not match the mapped values in Figure 3, the reported savings would be refuted.","tokens_in":9045,"feed_emoji":"📡","tokens_out":16968,"duration_ms":140163,"temperature":0.7,"pith_summary":"This paper proposes a coordinated hybrid precoding scheme for downlink massive MIMO in heterogeneous networks, where several base stations jointly serve users and deliberately shape inter-cell interference to be constructive — pushing each received symbol deeper into its correct decision region — rather than trying to cancel it. Hybrid precoding uses fewer radio-frequency chains than antennas, saving hardware cost, but normally costs extra transmit power; the paper's optimization minimizes total transmit power subject to quality-of-service and per-base-station power constraints, solved by a three-stage procedure of RF-chain assignment, constant-modulus analog precoding, and convex digital precoding. Simulation in a three-base-station network with 64 users shows that the coordinated constructive-interference design meets a given symbol error rate with lower transmit power than coordinated zero-forcing hybrid precoding (which cancels interference) and than uncoordinated constructive-interference precoding. It also shows a lower backhaul coordination overhead for the constructive-interference approach. The practical interest is energy efficiency in dense 5G networks without full-digital front ends.","feed_headline":"Exploiting interference cuts transmit power in coordinated networks","feed_subtitle":"At the same symbol error rate, it needs less transmit power than zero-forcing or uncoordinated hybrid precoding.","key_machinery":"The load-bearing object is the constructive-interference (CI) region of the PSK constellation: for each user, the received signal (after rotation by the conjugate of the intended symbol) must satisfy the cone constraint $\\left|\\operatorname{Im}(\\cdot)\\right| \\le (\\operatorname{Re}(\\cdot) - \\gamma_k) \\tan\\theta$, where $\\theta = \\pi/M$ and $\\gamma_k$ is the threshold margin controlling QoS. These constraints, together with the bilinear coupling of constant-modulus analog and digital precoders, make the joint problem nonconvex. The paper's mechanism for handling this is a three-stage decomposition: first a mixed-integer linear program assigns each RF chain to a user (maximizing total channel gain with fairness), then analog precoders are fixed as either phase-conjugated channel responses (continuous case) or codebook beams (codebook case), and finally the remaining digital precoding problem is convex and solved with standard tools. The CI constraints are what convert interference from an obstacle into a resource.","core_discovery":"The central claim is that inter-base-station interference, normally a nuisance in heterogeneous networks, can be exploited as a useful signal component if the precoders across base stations are designed jointly. By steering each user's received symbol into its constructive-interference region — the part of the PSK constellation cone that lies away from the decision boundaries — the coordinated hybrid precoder can fulfill quality-of-service constraints with less total transmit power than conventional zero-forcing hybrid precoding. The paper demonstrates this in simulation for a macro base station with 64 antennas and two pico base stations with 32 antennas each, serving 64 QPSK users, and additionally shows that the coordinated constructive-interference scheme reduces the backhaul load because it exchanges only $R_g$ digital coefficients per base station per symbol instead of $R_g \\times K$. The performance gain is reported empirically, not supported by an analytic performance bound.","pith_inferences":["If the power savings persist under imperfect channel state information, coordinated CI hybrid precoding could let operators run dense heterogeneous networks with cheaper, lower-power small cells; the paper assumes perfect CSI and does not test this.","The paper's symbol error rate values are produced through an empirical signal-quality-to-error-rate mapping from a single-cell study rather than direct symbol-level simulation; reproducing the comparison with end-to-end modulation and detection would test whether the reported savings are an artifact of that mapping.","The MILP-based RF-chain assignment could become a computational bottleneck in networks with many more base stations or users; a greedy or distributed assignment would be a natural testable extension.","The coordinated CI idea could combine with prior work on phase-error robustness to address hardware imperfections in the analog phase shifters."],"forward_implications":["Coordinated CI hybrid precoding can meet the same symbol error rate as coordinated zero-forcing hybrid precoding while using less total transmit power, according to the paper's simulations.","The three-stage decomposition keeps the computation practical: the analog and RF-chain assignment stages are handled separately from the convex digital precoding stage.","Codebook-based analog precoding, which avoids full-resolution phase shifters, works within the same coordinated framework and trades a modest performance loss for cheaper hardware.","The backhaul overhead of coordination is lower than for zero-forcing, because each base station receives $R_g$ digital coefficients per symbol rather than $R_g \\times K$.","The CI formulation is stated for PSK symbols but the paper indicates it extends to other modulation formats."],"supporting_citations":[{"why":"Establishes that known multiuser interference can be harnessed as useful signal power, providing the conceptual basis of CI-region constraints.","marker":"[22]"},{"why":"Provides the constructive-interference hybrid precoding approach and the phase-conjugation analog beamforming used in the continuous-valued case.","marker":"[27]"},{"why":"Supplies the single-cell CI hybrid precoding scheme and the empirical signal-quality-to-error-rate relationship used to produce the SER curves in Figure 3.","marker":"[28]"},{"why":"Defines the low-complexity zero-forcing hybrid precoding baseline that the proposed scheme is compared against.","marker":"[31]"},{"why":"The convex and mixed-integer solver used to compute the digital precoders and the RF-chain assignments.","marker":"[32]"},{"why":"Provides the macro and pico path-loss models used in the three-base-station simulation setup.","marker":"[35]"},{"why":"Defines the SINR-based user association rule used by the uncoordinated CI baseline.","marker":"[36]"}],"fun_headline_variants":["Interference turned asset: coordinated hybrid precoding cuts power","HetNet precoding exploits interference, shrinks transmit power","Coordinated hybrid precoding: interference as a power-saving tool","Joint precoding turns interference into energy efficiency","Less power for same QoS via interference-exploiting coordination"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The comparison assumes that the measured relationship between signal quality and symbol error rate from an earlier single-cell experiment holds unchanged for the coordinated multi-base-station scheme and for the zero-forcing baselines.","fun_headline_variants_meta":{"raw":{"variants":["Interference turned asset: coordinated hybrid precoding cuts power","HetNet precoding exploits interference, shrinks transmit power","Coordinated hybrid precoding: interference as a power-saving tool","Joint precoding turns interference into energy efficiency","Less power for same QoS via interference-exploiting coordination"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000286,"raw_usage":{"total_tokens":1626,"prompt_tokens":834,"completion_tokens":792,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":450,"completion_tokens_details":{"reasoning_tokens":713}},"tokens_in":450,"tokens_out":792,"duration_ms":8480,"temperature":1.0,"reasoning_tokens":713,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T14:17:16.905346+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Perform end-to-end Monte Carlo simulation in the same three-base-station heterogeneous network (macro with 64 antennas, two picos with 32 antennas, 64 QPSK users), transmitting actual symbols through the proposed coordinated CI hybrid precoder and the coordinated zero-forcing baseline, detecting with a maximum-likelihood rule, and plotting symbol error rate against transmit power; if zero-forcing achieves the same symbol error rate at lower or equal power than the CI scheme, or if the CI scheme's measured symbol error rate does not match the mapped values in Figure 3, the reported savings would be refuted.","supporting_citations":[{"cited_title":"Exploiting known interference as green signal power for downlink beamforming optimization,","cited_arxiv_id":null,"evidence_quote":"Establishes that known multiuser interference can be harnessed as useful signal power, providing the conceptual basis of CI-region constraints."},{"cited_title":"Analog beam- former design for interference exploitation based hybrid beam- forming,","cited_arxiv_id":null,"evidence_quote":"Provides the constructive-interference hybrid precoding approach and the phase-conjugation analog beamforming used in the continuous-valued case."},{"cited_title":"Interference exploitation-based hybrid precoding with robustness against phase errors,","cited_arxiv_id":null,"evidence_quote":"Supplies the single-cell CI hybrid precoding scheme and the empirical signal-quality-to-error-rate relationship used to produce the SER curves in Figure 3."},{"cited_title":"Low-complexity hybrid pre- coding in massive multiuser MIMO systems,","cited_arxiv_id":null,"evidence_quote":"Defines the low-complexity zero-forcing hybrid precoding baseline that the proposed scheme is compared against."},{"cited_title":"Mobility management challenges in 3GPP heterogeneous networks,","cited_arxiv_id":null,"evidence_quote":"Provides the macro and pico path-loss models used in the three-base-station simulation setup."},{"cited_title":"User association for load balancing in het- erogeneous cellular networks,","cited_arxiv_id":null,"evidence_quote":"Defines the SINR-based user association rule used by the uncoordinated CI baseline."}],"review_version":1}