Pith. sign in

REVIEW 3 major objections 6 minor 15 references

Microwave Linear Analog Computers (MiLACs) for Communications: Opportunities and Challenges

T0 review · 3 major / 6 minor · reviewed 2026-08-01 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Fully interconnected tunable-impedance microwave networks compute LMMSE estimation and matrix inversion at RF, replacing cubic digital complexity with quadratic parameter setting and cutting RF chains to one per data stream.

desk verdict A clearly written review/position paper that repackages the authors' own MiLAC results; the central complexity claims are all cited, not derived, and hinge on an ideal impedance-network model that remains unvalidated here. read the letter →

arxiv 2607.22509 v1 pith:YLWGB4UE submitted 2026-07-24 cs.IT eess.SPmath.IT

Matteo Nerini , Bruno Clerckx This is my paper
classification cs.ITeess.SPmath.IT
keywords analogcomputingmicrowavelinearcomputer(MiLAC)matrixinversionLMMSEestimationbeamformingMIMOzero-forcingtunableimpedancenetworks
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper argues that a reconfigurable network of linear microwave components—a MiLAC—can perform real matrix algebra on communication signals as they propagate, rather than in a digital baseband processor. Its central claim is that a fully interconnected network of tunable impedances has an input-output map equal to the regularized pseudo-inverse H^H(HH^H+λI)^{-1}, so LMMSE estimation and zero-forcing beamforming can be offloaded to radio-frequency hardware. Because the impedance values are linear combinations of the matrix entries, configuring the network costs O(MN) or O(N^2) operations, whereas digital LMMSE and matrix inversion cost O(MN^2) and O(N^3). The paper then reviews how MiLAC-aided transmitters and receivers can match fully digital beamforming with only as many RF chains as data streams, use low-resolution converters, and remove per-symbol matrix-vector products.

What carries the argument

The central object is the fully interconnected tunable-impedance microwave network, a MiLAC in which each port is coupled to all others and to ground through varactors or PIN diodes. Its defining property is structural nonlinearity: the network is linear in the signals, but the implemented transfer matrix is a nonlinear function of the tunable impedance values. Carrying the argument is the identity v = H^H (H H^H + λI)^{-1} u, which the network realizes by setting the impedances to linear combinations of H and λ. This identity turns LMMSE estimation and matrix inversion into O(MN)/O(N^2) parameter-setting tasks, and it is what makes the claimed cubic-to-quadratic complexity reduction possibl

What would settle it

Measure the scattering matrix of a fabricated N-port fully interconnected tunable-impedance network after setting impedances for a known H and λ, and compare it with H^H(HH^H+λI)^{-1}; the central claim fails if the discrepancy grows with N or if the time or energy required to reconfigure the components scales faster than N^2.

Watch

Extended reading notes

Core claim

On its own terms, the paper establishes that the class of linear microwave networks is not limited to fixed matrix multiplication. The load-bearing example is a network in which every port is connected to every other port and to ground through tunable impedances; although the network is linear in the signals, its output depends nonlinearly on the impedance settings. By choosing the impedances as linear combinations of a channel matrix H and a regularization scalar λ, the network's input-output map becomes v = H^H (H H^H + λI)^{-1} u—the LMMSE estimator. The same structure computes matrix inversion. This transfers the expensive pseudo-inverse step from the digital domain, where it costs O(N^3

Load-bearing premise

The central assumption is that a real network of tunable microwave impedances can be configured quickly and accurately enough to behave exactly like the regularized matrix inverse of the channel, so that only the linear-combination computation of the impedance values remains digital.

Editorial extensions

If this is right

  • Regularized zero-forcing beamforming and MMSE combining can be computed at radio frequency, with the per-coherence-time cost dropping from O(N_T N_S^2) to O(N_T N_S) for parameter setting.
  • Per-symbol matrix-vector precoding and combining becomes a single analog propagation step, eliminating the symbol-by-symbol O(N_T N_S) digital operation.
  • A lossless reciprocal MiLAC-aided transmitter or receiver matches fully digital spectral efficiency in single-user MIMO using only as many RF chains as data streams; hybrid digital-MiLAC and two-layer MiLAC extend this to multiuser systems.
  • DAC and ADC resolutions can be reduced to the number of levels needed for the modulation constellation (e.g., 1 bit for 4-QAM) in architectures where beamforming is purely analog.
  • The main hardware cost is the number of tunable impedances, which grows quadratically with ports; the paper reports that reduced-complexity MiLAC architectures can still be capacity-achieving in single-user systems.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper leaves implicit that the same network could compute regularized inverses for any matrix encoded as impedances, which extends beyond MIMO beamforming to other signal-processing tasks such as solving linear systems or MMSE channel estimation.
  • A caveat not developed in the paper is that the end-to-end advantage depends on reconfiguration cost; if tuning varactors or PIN diodes requires per-element calibration, lookup, or power that grows faster than O(N^2), the quadratic claim overstates the practical gain.
  • Because the identity is narrowband, the frequency response of the network limits how much bandwidth can share one configuration; wideband operation would require either frequency-flat designs or frequency-dependent impedance setting, an open direction the paper flags.
  • Structural nonlinearity suggests other nonlinear-in-parameters operations might be synthesized, including analog layers with parameter-dependent linear maps; the paper notes that a purely linear network cannot implement a true neural network without nonlinear devices.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. This paper is a review/tutorial introducing Microwave Linear Analog Computers (MiLACs), defined as linear microwave networks that perform matrix operations directly through wave propagation. It develops a narrowband model v=W(h)u, illustrates six implementations (wireless channel, RIS, SIM, microstrip circuits, phase-shifter networks, and fully interconnected tunable-impedance networks), and claims—citing ref. [2]—that a fully interconnected tunable-impedance MiLAC can realize LMMSE estimation and matrix inversion with O(MN)/O(N^2) parameter-setting complexity rather than the digital O(MN^2)/O(N^3). It then reviews MiLAC-aided transmitter/receiver architectures, including hybrid digital-MiLAC and two-layer MiLAC designs, and claims that these can match fully digital beamforming with few RF chains, lower DAC/ADC resolution, and reduced computational complexity. The final section lists open challenges: hardware impairments, channel estimation, wideband operation, and low-complexity architectures.

Significance. If the underlying hardware premise is accepted, the claimed quadratic-complexity matrix inversion and the reduction in RF chains and converter resolution would be genuinely significant for massive/gigantic MIMO. The paper is clearly written and provides a useful taxonomy of analog-computing structures, and it cites work beyond the authors' own (e.g., [9], [12]). It also explicitly acknowledges hardware non-idealities in Section V.A. However, the central complexity and optimality results are inherited from prior work (mainly [2], [8], [10], [11]) and are neither derived nor experimentally validated in this manuscript. The significance is therefore conditional on an ideal hardware model whose practical validity is not established here.

major comments (3)
  1. [II-B, II-C, Table I] The load-bearing claim that a fully interconnected tunable-impedance network computes v=H^H(HH^H+λI)^{-1}u, and that this can be configured with only O(MN)/O(N^2) work, is asserted as established fact via citation to [2]. No construction of H and λ from the impedance values, no existence conditions, and no approximation/error bounds are given in this manuscript. Since the abstract and Table I present this as a definitive result and the rest of the paper builds on it, the central premise is not verifiable from the paper itself. Please either include a derivation (or a precise reference to a derivation with the assumptions made explicit) and a statement of the idealization under which it holds, or clearly mark this as a known result from [2] and condition the headline claims accordingly.
  2. [II-C, Table I, V.D] The complexity comparison in Table I counts the O(MN)/O(N^2) arithmetic needed to compute impedance values as linear combinations of the matrix entries, while treating the analog computation as O(1). However, the actual cost of reconfiguring the network—digital-to-analog conversion for the tunable components, settling time, calibration, and control overhead—is not modeled. Section V.D itself notes that the fully interconnected architecture has O(N^2) tunable components. If reconfiguration is not negligible, the claimed 'quadratic rather than cubic' advantage may not survive in practice. This is not merely a presentation issue: it affects the validity of the central complexity comparison. The paper should either provide a model of the reconfiguration cost or explicitly restrict the claim to the ideal case and flag the missing overhead.
  3. [V.A vs Abstract/III] The abstract and Section III make unconditional statements that MiLAC-aided architectures 'achieve the same performance as fully digital beamforming' with fewer RF chains and lower DAC/ADC resolution. Section V.A then states that hardware non-idealities (discrete tunable values, insertion losses, mutual coupling, impedance matching) 'must be accounted for to assess the true performance.' The paper does not quantify how these impairments affect the earlier optimality and complexity claims, leaving the headline benefits without a stated validity regime. A review paper should make the ideal-vs-nonideal boundary explicit, e.g., by stating that all performance/complexity claims are for ideal lossless reciprocal networks and by summarizing the known impact of each non-ideality, even at a qualitative level.
minor comments (6)
  1. [Abstract, I] The abstract and introduction say 'we show' for results that are reviewed from prior work. For a survey paper, 'we review and discuss' would be more accurate.
  2. [II-A] The term 'structural nonlinearity' is introduced but not defined or cited. Provide a precise definition and a reference.
  3. [II-C, Table I] The 'O(1)' entry for analog matrix-vector multiplication is potentially confusing. Clarify that it means a single propagation step independent of M,N under the adopted complexity model, not a conventional arithmetic operation count.
  4. [III-A] The statement that a lossless reciprocal MiLAC 'can only implement semi-unitary transformations' is given intuitively. Add a one-line argument or an explicit reference to [8] so the reader can follow the limitation.
  5. [IV-B] The claim that 4-QAM requires only 1-bit DACs is valid for the baseband symbol values, but the RF chain and upconversion may impose additional linearity/filtering requirements. State this qualification to avoid overgeneralization.
  6. [References] Reference [1] is listed as a preprint/working paper. Mark it as such in the text as well, and ensure the paper's survey scope is clearly distinguished from the new contributions of the cited works.

Circularity Check

2 steps flagged · score 6.0 of 10

Central quadratic-inversion and digital-beamforming-equivalence claims are carried by same-author citations [2], [8], [10], [11] rather than derived or independently validated in this paper.

  1. self citation load bearing [Section II-B, Fig. 2(f); Section II-C, Table I]
    "In particular, it is possible to identify a matrix H ... and a scalar λ, both given by linear combinations of the tunable parameters, such that the computed transformation is v = H^H (HH^H + λI)^{-1} u [2]. ... It has been shown in [2] that such a MiLAC can compute both the LMMSE estimator and matrix inversion ... as shown in [2]."

    The paper's central claim—that a tunable-impedance MiLAC can realize LMMSE estimation and matrix inversion with O(MN)/O(N^2) parameter-setting cost—is not derived here. The construction and the complexity scaling are attributed entirely to [2], which is authored by the present authors. Table I inherits the assertion that the impedance values are linear combinations of H and λ without any proof or independent evidence in this manuscript. Thus the headline result reduces to a load-bearing same-author citation.

  2. self citation load bearing [Section III-A and III-B]
    "Under these constraints, it has been shown analytically that a MiLAC-aided transmitter or receiver can achieve the same performance as fully digital beamforming in single-user MIMO systems [8]. ... As a result, hybrid digital-MiLAC beamforming can achieve optimal performance in general wireless communication scenarios using only as many RF chains as data streams [10]. ... In this way, this two-layer MiLAC architecture also achieves optimal performance ... [11]."

    The paper's key system-level benefits—matching fully digital beamforming while using only as many RF chains as data streams—are asserted with citations [8], [10], [11], all of which include the present authors. The manuscript offers only an intuitive SVD explanation and no proof or external validation. These optimality and RF-chain-savings claims are therefore supported by a same-author citation chain rather than by derivations contained in or independently verified by this paper.

full rationale

This is a review/position paper, so some reliance on prior work is expected and not itself circular. I found no equation-level circularity: the LMMSE identity v = H^H(HH^H+λI)^{-1}u is not defined into existence, and no fitted parameter is relabeled as a prediction. However, the two load-bearing claims of the paper—quadratic-complexity matrix inversion/pseudo-inversion (Section II) and fully-digital-equivalent beamforming with minimal RF chains (Section III)—are not derived in the manuscript. They are attributed to [2], [8], [10], and [11], all authored or co-authored by the present authors, and the paper supplies only an SVD sketch as intuition. Section V.A further acknowledges that practical hardware effects (discrete tuning, insertion loss, mutual coupling) are unmodeled, so the ideal impedance-network mapping that underlies the central claims remains an unvalidated premise imported from same-author work. Because the underlying algebra is not circular by construction and the paper is transparently a survey, the circularity is moderate rather than definitional.

Assumptions & free parameters 0 free parameters · 5 assumptions · 0 invented entities

No free parameters are fitted to data; the paper reports no experiments. The load-bearing structure is a set of modeling choices: narrowband linear multiport model, losslessness/reciprocity, the algebraic identity v = H^H (H H^H + λI)^{-1} u attributed to [2], and a complexity model that counts only the digital computation of tunable parameters. The main unverified premise is that arbitrary H and λ can be mapped to physical impedance values with negligible error and overhead.

assumptions (5)
  • domain assumption A narrowband linear microwave network is a multi-port whose output is v = W(h)u.
    Section II-A; relies on linearity and time-invariance of transmission lines, phase shifters, and impedances.
  • domain assumption Passive tunable impedance components are lossless and reciprocal in the idealized model.
    Section III-A; says 'under the standard microwave constraints of losslessness and reciprocity'; used for semi-unitary optimality.
  • ad hoc to paper For a fully interconnected tunable impedance network, there exist H and λ (linear combinations of tunable parameters) such that v = H^H (H H^H + λI)^{-1} u.
    Section II-B, attributed to [2]; not derived in this paper; this is the algebraic core of the O(N^2) inversion claim.
  • ad hoc to paper The digital cost of setting tunable parameters equals the cost of computing linear combinations O(MN)/O(N^2), ignoring analog control/reconfiguration overhead.
    Section II-C and Table I; if physical reconfiguration cost scales or dominates, the complexity advantage is overstated.
  • domain assumption Narrowband operation holds; wideband effects are deferred.
    Section II-A and V-C; all complexity and optimality claims are single-frequency.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Microwave Linear Analog Computers (MiLACs) for Communications: Opportunities and Challenges." pith.science (2026). https://pith.science/paper/YLWGB4UE

@misc{pith2026260722509,
  author       = {Pith},
  title        = {Pith review of: Microwave Linear Analog Computers (MiLACs) for Communications: Opportunities and Challenges},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YLWGB4UE}},
  note         = {Machine review of arXiv:2607.22509}
}
read the original abstract

Future wireless systems will require ever larger antenna arrays and heavier signal processing, making conventional digital multiple-input multiple-output (MIMO) architectures difficult to scale. In this paper, we show that a possible solution is to offload part of the processing from the digital to the analog domain. This can be done through linear microwave networks designed to compute directly using the communication signals at radio frequency (RF). These networks, denoted as microwave linear analog computers (MiLACs), can perform useful matrix operations instantly through wave propagation. Remarkably, although MiLACs are linear, the output signals can depend nonlinearly on the tunable parameters of the network, enabling the computation of operations beyond simple linear transforms. In particular, MiLACs can realize matrix inversion and pseudo-inversion with complexity scaling quadratically with matrix size, rather than cubically, which is essential in zero-forcing beamforming. We then review how MiLAC-aided MIMO architectures can reduce the number of RF chains, relax the resolution requirements on digital-to-analog converters (DACs) and analog-to-digital converters (ADCs), and decrease the beamforming complexity. We finally discuss the main challenges related to MiLAC and promising directions for future research.

Figures

Figures reproduced from arXiv: 2607.22509 by the authors.

Figure 1
Figure 1. A MiLAC whose linear microwave network has [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Six examples of linear microwave networks that can implement a MiLAC: (a) a wireless channel, (b) a RIS-aided wireless channel, (c) a SIM-aided [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. A MiLAC fabricated in microstrip technology that computes the [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: (a) A MiLAC-aided transmitter precoding the symbols in the analog domain, and (b) a MiLAC-aided receiver combining the signals at the receiving [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: (a) A hybrid digital-MiLAC-aided transmitter, and (b) a two-layer MiLAC-aided transmitter. With lossless and reciprocal MiLACs, these architectures [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

15 extracted references · 3 linked inside Pith

  1. [2]

    Analog computing for signal processing and communications – Part I: Computing with microwave networks,

    M. Nerini and B. Clerckx, “Analog computing for signal processing and communications – Part I: Computing with microwave networks,”IEEE Trans. Signal Process., vol. 73, pp. 5183–5197, 2025

    2025

  2. [3]

    Analog computing for signal processing and communications – Part II: Toward gigantic MIMO beamforming,

    M. Nerini and B. Clerckx, “Analog computing for signal processing and communications – Part II: Toward gigantic MIMO beamforming,”IEEE Trans. Signal Process., vol. 73, pp. 5198–5212, 2025

    2025

  3. [8]

    Capacity of MIMO systems aided by microwave linear analog computers (MiLACs),

    M. Nerini and B. Clerckx, “Capacity of MIMO systems aided by microwave linear analog computers (MiLACs),”arXiv preprint arXiv:2506.05983, 2025

    arXiv 2025

  4. [10]

    Microwave linear analog computer (MiLAC)-aided multiuser MISO: Fundamental limits and beamforming design,

    Z. Wu, M. Nerini, and B. Clerckx, “Microwave linear analog computer (MiLAC)-aided multiuser MISO: Fundamental limits and beamforming design,”arXiv preprint arXiv:2601.10060, 2026

    arXiv 2026

  5. [11]

    Two-layer microwave linear analog computer (MiLAC)-aided multi-user MISO networks,

    X. Zhou, T. Fang, Y . Mao, and B. Clerckx, “Two-layer microwave linear analog computer (MiLAC)-aided multi-user MISO networks,”arXiv preprint arXiv:2604.24303, 2026

    arXiv 2026

  6. [9]

    On the performance of lossless reciprocal MiLAC architectures in multi-user networks,

    T. Fang, X. Zhou, and Y . Mao, “On the performance of lossless reciprocal MiLAC architectures in multi-user networks,”IEEE Wireless Commun. Lett., vol. 15, pp. 2609–2613, 2026

    2026

  7. [12]

    Microwave linear analog computers aided multiuser communication: General impedance matching and pre- coding optimization,

    R. Xiong, J. Xu, and R. Zhang, “Microwave linear analog computers aided multiuser communication: General impedance matching and pre- coding optimization,”arXiv preprint arXiv:2606.04532, 2026

    arXiv 2026

  8. [1]

    Programmable wave-domain computing in wireless communications,

    P. del Hougne, M. D. Renzo, A. Al `u, T. Jun Cui, Y . Eldar, N. Engheta, W. Hu, and A. Ozcan, “Programmable wave-domain computing in wireless communications,” Feb. 2026, working paper or preprint. [Online]. Available: https://hal.science/hal-05487878

    2026

Show all 15 references
  1. [4]

    Intelligent reflecting surface-aided wireless communications: A tutorial,

    Q. Wu, S. Zhang, B. Zheng, C. You, and R. Zhang, “Intelligent reflecting surface-aided wireless communications: A tutorial,”IEEE Trans. Commun., vol. 69, no. 5, pp. 3313–3351, 2021

    2021

  2. [5]

    Stacked intelligent metasurfaces for efficient holographic MIMO communications in 6G,

    J. An, C. Xu, D. W. K. Ng, G. C. Alexandropoulos, C. Huang, C. Yuen, and L. Hanzo, “Stacked intelligent metasurfaces for efficient holographic MIMO communications in 6G,”IEEE J. Sel. Areas Commun., vol. 41, no. 8, pp. 2380–2396, 2023

    2023

  3. [6]

    Analog computing with hybrid couplers and phase shifters,

    M. Nerini, X. Liu, and B. Clerckx, “Analog computing with hybrid couplers and phase shifters,”arXiv preprint arXiv:2603.24604, 2026

    2026

  4. [7]

    Hybrid digital and analog beamforming design for large-scale antenna arrays,

    F. Sohrabi and W. Yu, “Hybrid digital and analog beamforming design for large-scale antenna arrays,”IEEE J. Sel. Top. Signal Process., vol. 10, no. 3, pp. 501–513, 2016

    2016

  5. [13]

    Channel estimation in MIMO systems aided by microwave linear analog computers (MiLACs),

    Q. Zhang, M. Nerini, and B. Clerckx, “Channel estimation in MIMO systems aided by microwave linear analog computers (MiLACs),”arXiv preprint arXiv:2601.11438, 2026

    2026

  6. [14]

    MIMO systems aided by microwave linear analog computers: Capacity-achieving architectures with reduced circuit complexity,

    M. Nerini and B. Clerckx, “MIMO systems aided by microwave linear analog computers: Capacity-achieving architectures with reduced circuit complexity,”IEEE Trans. Wireless Commun., vol. 25, pp. 14 597–14 610, 2026

    2026

  7. [15]

    Beamforming design for stem-connected microwave linear analog computer (MiLAC)- aided multiuser MISO downlinks,

    Y . Zhang, Z. Wu, B. Clerckx, and T. Y . Al-Naffouri, “Beamforming design for stem-connected microwave linear analog computer (MiLAC)- aided multiuser MISO downlinks,”arXiv preprint arXiv:2606.14499, 2026. BIOGRAPHIES Matteo Neriniis a Postdoctoral Research Associate in the De...

    2026

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.