{"id":"db723f25-2ee8-4ec5-b2ac-526b15d75f37","arxiv_id":"2412.02023","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":8,"one_line_summary":"A modified Flower Pollination Algorithm with entropy-based diversity control is proposed for joint channel estimation and symbol detection in synchronous DS/CDMA, and simulations show improved BER over several existing detectors.","lead":"This paper modifies a nature-inspired optimization method, the Flower Pollination Algorithm, to jointly estimate channels and decode user symbols in DS/CDMA wireless systems. Simulations suggest the modified algorithm achieves lower bit error rates than several conventional and nature-inspired detectors, especially under interference and near-far conditions.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"In Sec. 3.2, Eq. (16) defines Shannon entropy over normalized likelihood values that can be negative, making the diversity-control step undefined for valid populations.","rationale":"The reader identified the entropy-control mechanism as the weakest assumption, and I agree that this is the load-bearing part of the paper. However, the reader's formulation focused on the under-specified \"first or last iterations\" rule, whereas the more fundamental problem is that Eq. (16) is mathematically ill-defined for the fitness function defined in Eq. (7). The log-likelihood can take negative values, so the normalized fitness values used in the Shannon entropy can be negative, making the logarithm undefined. This is an internal inconsistency, not a matter of tuning or missing parameters. The paper's reported results may still be reproducible with a different fitness scaling (for example, using exp(L_i) or rank-based values), but the text does not specify such a transformation, and no code is provided. Therefore, as written, the proposed algorithm cannot be executed, and the claimed BER advantage cannot be attributed to the entropy-diversity mechanism. I recommend an unverdict until the authors provide the exact fitness transformation and, ideally, source code; if the entropy computation cannot be repaired, the paper should be rejected. The paper does have strengths: a clear channel model, comparisons to several detectors, and statistical significance tests, so this is not a dismissal of the simulation effort, but a precise reproducibility obstacle at the core of the claimed novelty.","tokens_in":22099,"tokens_out":9712,"duration_ms":99550,"concrete_test":"Analytical check: evaluate Eq. (7) for the U=1, BPSK, noiseless case with a=1, E=1, R=1 and population {+1,-1}; the normalized fitnesses are -0.5 and 1.5, so Eq. (16) is undefined. If the authors specify a different fitness scaling, re-run the Section 4.1 experiment (SNR=11 dB, U=10) using that scaling; if the Table 1 BER for FPA changes by more than one order of magnitude, the reported result was an artifact of an unspecified transformation.","verdict_should_be":"UNVERDICTED","load_bearing_attack":"The paper's central novelty is the entropy-driven adjustment of Pc (Sec. 3.2). The entropy in Eq. (16) is computed from normalized fitness values L_i^* = L_i / sum_j L_j, where L_i is the log-likelihood of Eq. (7). This log-likelihood is not nonnegative: for a single-user BPSK noiseless channel with a=1, E=1, and true symbol d=+1, Eq. (7) gives L(+1)=1 and L(-1)=-3; the normalized values are -0.5 and 1.5, so log(L_i^*) is undefined. In a 25-flower population initialized with random symbols, mixed-sign likelihoods are the norm, not the exception. Consequently Eq. (16) is not a valid Shannon entropy for the fitness function used, and the proposed Pc update rule cannot be executed as described. The paper also leaves the direction and magnitude of Pc changes unquantified (\"Modifications of Pc are smaller as iteration t increases\" with no formula, and the \"first or last iterations\" disambiguation in Sec. 3.2 is informal). Since the reported BER advantage is attributed to this diversity control, the central claim is not reproducible from the text alone.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a modified Flower Pollination Algorithm (FPA) for joint symbol detection and flat-fading channel estimation in synchronous DS/CDMA systems. The main novelties are an entropy-based adaptation of the probability of change Pc (Sec. 3.2) and a split encoding in which global and local pollination act differently on the symbol part and the channel-coefficient part of each candidate solution. The authors evaluate the resulting detector by Monte Carlo simulations of BER versus SNR, BER versus number of users, channel-estimation MSE, and near-far performance, comparing against matched filter, decorrelator, MMSE-RLS, Mahalanobis-RBF, standard GA, tabu search, simulated quenching, and several Bayesian detectors, with Kruskal-Wallis and Friedman tests for statistical significance. The central claim is that the proposed FPA offers competitive or lower BER under difficult conditions (many users, low SNR, near-far effects) while requiring less computation than conventional methods with similar performance.","tokens_in":22457,"tokens_out":5876,"duration_ms":55812,"significance":"If the algorithm were fully and reproducibly specified, the claimed results would be of practical interest for DS/CDMA multiuser detection, particularly the near-far resistance and the ability to adapt exploration/exploitation through population-diversity monitoring. The paper has clear strengths: the BER and MSE results are direct simulation outcomes rather than circular derivations; the authors compare against several strong baselines, including their own previous Bayesian detectors; and they report statistical significance tests and make an explicit effort to equalize computational load in terms of fitness-function evaluations. These features make the empirical comparison credible in broad outline. However, the significance is currently conditional on correcting a load-bearing technical defect: the entropy measure in Eq. (16) is undefined for the log-likelihood fitness in Eq. (7), and the adaptive Pc rule in Sec. 3.2 is under-specified. These issues prevent exact reproduction and make it impossible to isolate the contribution of the paper's central novelty.","major_comments":[{"comment":"The Shannon entropy in Eq. (16) uses normalized fitness values L_i^*[k] = L_i[k] / (sum_j L_j[k]) as probabilities, but the log-likelihood L in Eq. (7) is not nonnegative. For a single-user BPSK noiseless channel with a=1, E=1 and true symbol d=+1, Eq. (7) gives L(+1)=1 and L(-1)=-3; after normalization both signs can occur, so log L_i^* is undefined for typical populations, including the 25-flower random initializations described in Sec. 4. The entropy-controlled Pc update therefore cannot be executed as written, and the central novelty of the paper is not reproducible from the text. Please redefine the diversity measure over a nonnegative fitness mapping (for example, exponentiating L or using likelihoods rather than log-likelihoods) or justify the current formula.","section":"Sec. 3.2, Eq. (16)"},{"comment":"The adjustment rule for Pc is under-specified. The two bullets state only qualitative directions (higher entropy increases Pc, lower entropy decreases Pc), and the text adds that modifications are smaller as iteration t increases and are applied every MaxIter/10 iterations, but it gives no formula for the step size, no bounds on Pc, and no quantitative criterion for deciding whether the algorithm is in the 'first or last iterations'. Because the BER advantage over the standard FPA is attributed to this adaptive mechanism, the missing rule prevents exact reproduction and also prevents isolating the entropy control's contribution. Please provide an explicit update equation for Pc(t) or a table of the parameter settings used in the simulations.","section":"Sec. 3.2, step 6"},{"comment":"Step 1 of the algorithm description states that the best solution b* is the flower with the lowest fitness value, whereas Eq. (8) and step 7 define the solution as the one maximizing the log-likelihood L. Since the selection and replacement logic in step 5 depends on this objective, the two statements cannot both be correct. Please correct the description and state explicitly whether L is maximized or minimized in the simulations.","section":"Sec. 3.1, step 1"},{"comment":"The near-far experiment introduces an ISI channel H_UOI(z) = a0 + 0.471 z^{-1} + 0.063 z^{-2}, but the system model in Sec. 2 and the flower encoding in Sec. 3.1 assume a single flat-fading coefficient a_i(n) per user with no intersymbol interference. It is not explained how this multipath transfer function is incorporated into Eq. (7), the state vector in Eq. (9), or the matched-filter output z(n). As the near-far resistance claim rests on this experiment, please reconcile the channel model or clarify the extension used there.","section":"Sec. 4.5"}],"minor_comments":[{"comment":"The Standard GA row at SNR=10 dB reports 0.007 ± 0.019, which has a standard deviation larger than the mean and is inconsistent with the decreasing BER trend; please check this entry.","section":"Table 1"},{"comment":"The subscript i is used both for the flower index and for the elements d_i,t(u), which can be confused with user indices; consider denoting the flower index differently, for example x^(m)[t].","section":"Eq. (9)"},{"comment":"The caption mentions 'np: population size of the standard GA', but np is not defined in the text; please specify the GA population size and the number of generations used for the results in Fig. 3.","section":"Sec. 4.1, Fig. 3 caption"},{"comment":"The abbreviation 'MMSEE' appears in the abstract and in Sec. 4.3, while the text mainly uses 'MMSE' and 'MMSE-RLS'; please unify the terminology.","section":"General terminology"},{"comment":"The text says that 'the first coefficient (a0) varies in such a way that signal-to-interference ratio is -10 ≤ SIR ≤ 10 dB', but the horizontal axis of Fig. 7 is labeled 'Power difference (dB)'; please clarify the relationship between SIR and the plotted quantity.","section":"Sec. 4.5"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know this paper before recommending it: the claimed novelty—entropy-based diversity control—is not actually well-defined as written, and the simulation section does not support the 'higher symbol rates' claim in the abstract. That said, the application of FPA to joint channel estimation and symbol detection in DS/CDMA is new, and the paper does a thorough comparison job.\n\nThe new bit is applying Yang's FPA to the DS/CDMA equalization problem, splitting each flower into a symbol part and a fading-coefficient part, and adjusting the probability of global vs local pollination based on the Shannon entropy of the normalized fitness. The split encoding is sensible, and the simulation campaign is broad: against MF, decorrelator, MMSE-RLS, GA, TS, SQ, and a couple of Bayesian detectors, with Kruskal-Wallis and Friedman tests. The reported BER results are plausible, and the complexity comparison (Table 3) is useful.\n\nThe soft spot is not minor. Eq. (16) defines entropy over normalized fitness values L_i* = L_i / sum L_j, with L_i the log-likelihood from Eq. (7). Log-likelihoods can be negative. In a simple single-user noiseless case, L(+1)=1 and L(-1)=-3, so the normalized values are -0.5 and 1.5, and log(L_i*) is undefined. Mixed-sign likelihoods are typical in the initial random population, so the entropy-control step cannot be executed as described. The paper also leaves the Pc adjustment rule informal: 'we know if the algorithm is in the first or last iterations' is not a rule, and the step size is unquantified. This matters because the BER advantage over standard FPA is attributed to exactly this mechanism. Without a corrected definition (e.g., using probabilities or a positive transform), the algorithm is not reproducible from the text.\n\nTwo smaller issues: the near-far experiment in Sec. 4.5 quietly introduces an ISI channel (HUOI(z)) while the rest of the paper assumes flat fading, and the 'higher symbol rates' claim in the abstract and conclusions is never tested—no symbol-rate axis anywhere. The paper also relies heavily on the authors' own earlier Bayesian detectors, but those comparisons are legitimate.\n\nWho is this for? A reader working on metaheuristic MUD for conventional CDMA systems might get a useful baseline and a cautionary example of under-specified diversity control. It deserves a serious referee because the empirical work is substantial, but the entropy flaw is load-bearing and needs to be fixed before the results can be trusted. My recommendation: engage with it, send it to review, but flag the entropy issue early.","headline":"Entropy-based diversity control in the proposed FPA is not well-defined as written, undermining the central claim, but the application and simulations are substantial.","tokens_in":22901,"tokens_out":2957,"would_cite":false,"duration_ms":28034,"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":"A modified Flower Pollination Algorithm for joint symbol detection and channel estimation in synchronous DS/CDMA claims near-single-user bit error rates under near-far interference and lower runtime than a standard genetic algorithm.","keywords":["DS/CDMA","Flower Pollination Algorithm","multiuser detection","channel estimation","population diversity control","entropy-based adaptation","near-far effect","nature-inspired optimization"],"falsifier":"Re-implement the algorithm exactly as specified, with no hidden iteration-based rule for interpreting high entropy, and run the BER versus SNR experiment of Fig. 3: if the results differ materially from Table 1, the entropy-control mechanism is not fully specified. A sharper test: replace the entropy rule with a fixed $P_c = 0.35$ and a fixed schedule of shrinking $\\gamma_a$; if that simpler version matches the reported BER, the entropy monitoring is not what carries the performance.","tokens_in":21931,"feed_emoji":"🌸","tokens_out":6364,"duration_ms":57747,"temperature":0.7,"pith_summary":"This paper proposes a modified Flower Pollination Algorithm (FPA) for the joint estimation of channel fading coefficients and transmitted symbols in synchronous DS/CDMA systems. The authors claim the modified detector reaches bit error rates close to the single-user bound when the user of interest is weak, many interferers are active, or near-far effects are strong, conditions where matched filter, decorrelator, genetic algorithm, tabu search, and simulated quenching degrade. The key modification is monitoring the Shannon entropy of the population's fitness values and using it to adjust the probability of switching between global and local pollination, preserving diversity during the search. Compared at equal fitness evaluations, the proposed FPA reaches the same BER in about 45% of the time a standard genetic algorithm needs, and the reported differences are backed by Kruskal-Wallis and Friedman tests.","feed_headline":"FPA detector nears single-user bound under near-far CDMA interference","feed_subtitle":"Joint symbol and channel estimation with entropy-driven diversity reaches target BER in half the time of a standard genetic algorithm.","key_machinery":"The load-bearing mechanism is the entropy-controlled probability of change in the standard Flower Pollination Algorithm. Population diversity is monitored through the Shannon entropy of normalized flower fitnesses, $H(P[k]) = -\\sum_i L^*_i[k] \\log L^*_i[k]$; high entropy (similar fitnesses) raises $P_c$ to favor global pollination with a L\\'evy flight, and low entropy lowers $P_c$ to favor local pollination. The solution encoding is split into a symbol part $d(n)$ and a fading part $a(n)$, which receive different scale factors $\\gamma_d$ and $\\gamma_a$, with $\\gamma_a$ shrinking by 10% every $MaxIter/4$ iterations so the already-estimated channel coefficients are only fine-tuned while symbols are re-randomized each symbol period. Fitness is the log-likelihood expression from the DS/CDMA model, and $P_c$ adjustments are applied every $MaxIter/10$ iterations with smaller updates as iterations grow.","core_discovery":"On the paper's own terms, the central claim is that a Flower Pollination Algorithm whose exploration-exploitation balance is controlled by the entropy of fitness values is a practical near-far-resistant multiuser detector. Each candidate solution is a vector of the form $[d(n)|a(n)]$ containing all users' transmitted symbols and their flat-fading coefficients, and its fitness is the log-likelihood of the received matched-filter output. Global pollination, modeled as a L\\'evy flight toward the current best solution, is applied with a larger step to the symbol part and a smaller, decreasing step to the fading part, while local pollination uses random pairwise differences. The probability of change $P_c$, which chooses between global and local pollination, is increased when fitness entropy is high and decreased when it is low, so the population keeps exploring early and exploits late. The paper reports lower BER than the compared conventional and nature-inspired detectors under low SNR, many users, and near-far imbalance, with channel-estimation MSE near the single-user limit and no need for supervised initialization.","pith_inferences":["If the entropy-based diversity control is what carries the gain, the same control rule could be bolted onto other population metaheuristics (PSO, cuckoo search, cat swarm) for the same DS/CDMA problem; the paper does not test that transfer.","Because the ambiguity about early versus late high-entropy states is resolved only by knowing the iteration count, the adaptive $P_c$ may effectively reduce to an iteration-dependent schedule; a fixed schedule might reproduce most of the benefit.","The complexity comparison is in time-to-target-BER, so the practical \"less complexity\" claim should be checked with a direct operation count or hardware implementation, since entropy computation and population management add overhead not captured by fitness evaluations.","Extending the algorithm to asynchronous users or non-binary alphabets, which the paper lists as future work, is non-trivial because the model assumes perfect synchronization, BPSK, and symbol-rate processing; the near-far robustness shown here is for that narrower setting."],"forward_implications":["Under low SNR, many active users, or 4-15 dB near-far power imbalance, the entropy-controlled FPA keeps BER closer to the single-user bound than MF, decorrelator RLS, MMSE-RLS, Std-GA, TS, and SQ in the tested regimes.","At equal fitness-function evaluations, the FPA reaches the same target BER in about 45% of the time used by Std-GA and about two-thirds of the time used by TS or SQ.","The channel estimator part of the FPA reaches MSE near the single-user limit, comparable to a Bayesian detector, without requiring a supervised training period or burn-in.","Kruskal-Wallis tests with Bonferroni correction and a Friedman test report statistically significant performance differences favoring the FPA over Std-GA, TS, and SQ for most tested SNRs."],"supporting_citations":[{"why":"Supplies the base Flower Pollination Algorithm, including global and local pollination rules, the L\\'evy flight model, and the probability of change $P_c$ that the paper modifies.","marker":"(Yang, 2012)"},{"why":"Supplies the joint symbol detection and fading channel estimation formulation, including the maximization framework and the time-varying fading model that the paper adopts.","marker":"(Yen & Hanzo, 2001)"},{"why":"Supplies the log-likelihood expression used as the fitness function in Eq. (7), the objective that the FPA maximizes.","marker":"(Fawer & Aazhang, 1995)"},{"why":"Supplies the digital communication channel model, the single-user bound, and the definition of near-far and interference conditions used in the comparisons.","marker":"(Proakis, 1998)"},{"why":"Establishes that optimal maximum-likelihood multiuser detection has complexity exponential in the number of users, motivating the need for suboptimal detectors like the proposed FPA.","marker":"(Verdu, 1998)"},{"why":"Provides the matched filter and decorrelator linear detectors used as conventional baselines in the BER and near-far experiments.","marker":"(Lupas & Verdu, 1989)"},{"why":"Supplies the decorrelator RLS detector and the Kalman-based channel estimator used as baselines in both BER and MSE comparisons.","marker":"(Lim et al., 1998)"},{"why":"Supplies the MAP-GCGS and MAP-GS Bayesian schemes used as channel-estimation benchmarks that the FPA is compared against in the MSE plots.","marker":"(Huang & Djuric, 2002)"}],"fun_headline_variants":["Entropy-tuned flower pollination defeats near-far CDMA interference","Flower pollination with entropy control nears single-user bound","FPA with fitness entropy halves convergence time over GA in CDMA","Nature-inspired FPA matches single-user BER under near-far fading"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The advantage over the standard FPA depends on the claim that the algorithm can tell whether high fitness entropy means the population is still exploring or has converged, but the paper gives no formal rule for making that distinction.","fun_headline_variants_meta":{"raw":{"variants":["Entropy-tuned flower pollination defeats near-far CDMA interference","Flower pollination with entropy control nears single-user bound","FPA with fitness entropy halves convergence time over GA in CDMA","Nature-inspired FPA matches single-user BER under near-far fading"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000454,"raw_usage":{"total_tokens":2300,"prompt_tokens":981,"completion_tokens":1319,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":597,"completion_tokens_details":{"reasoning_tokens":1246}},"tokens_in":597,"tokens_out":1319,"duration_ms":9614,"temperature":1.0,"reasoning_tokens":1246,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T23:54:32.349467+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Re-implement the algorithm exactly as specified, with no hidden iteration-based rule for interpreting high entropy, and run the BER versus SNR experiment of Fig. 3: if the results differ materially from Table 1, the entropy-control mechanism is not fully specified. A sharper test: replace the entropy rule with a fixed $P_c = 0.35$ and a fixed schedule of shrinking $\\gamma_a$; if that simpler version matches the reported BER, the entropy monitoring is not what carries the performance.","supporting_citations":[{"cited_title":"( year 2012 )","cited_arxiv_id":null,"evidence_quote":"Supplies the base Flower Pollination Algorithm, including global and local pollination rules, the L\\'evy flight model, and the probability of change $P_c$ that the paper modifies."},{"cited_title":", & author Hanzo, L","cited_arxiv_id":null,"evidence_quote":"Supplies the joint symbol detection and fading channel estimation formulation, including the maximization framework and the time-varying fading model that the paper adopts."},{"cited_title":", & author Aazhang, B","cited_arxiv_id":null,"evidence_quote":"Supplies the log-likelihood expression used as the fitness function in Eq. (7), the objective that the FPA maximizes."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the digital communication channel model, the single-user bound, and the definition of near-far and interference conditions used in the comparisons."},{"cited_title":"( year 1998 )","cited_arxiv_id":null,"evidence_quote":"Establishes that optimal maximum-likelihood multiuser detection has complexity exponential in the number of users, motivating the need for suboptimal detectors like the proposed FPA."},{"cited_title":", & author Verdu, S","cited_arxiv_id":null,"evidence_quote":"Provides the matched filter and decorrelator linear detectors used as conventional baselines in the BER and near-far experiments."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the decorrelator RLS detector and the Kalman-based channel estimator used as baselines in both BER and MSE comparisons."},{"cited_title":", & author Djuric, P","cited_arxiv_id":null,"evidence_quote":"Supplies the MAP-GCGS and MAP-GS Bayesian schemes used as channel-estimation benchmarks that the FPA is compared against in the MSE plots."}],"review_version":1}