A Research Agenda for the Next Decade

Why a Research Agenda at the End of a Book

Most textbook chapters end with a summary; this one ends with a program. The reason is that MIMO research is not closing — the preceding five sections each identified problems that will plausibly occupy PhD students through 2035. The purpose of this final section is to state those problems in a form that is useful to someone deciding what to work on next.

The agenda is organized by theme rather than by chapter. A single research project will often touch more than one of the open problem areas. We close by identifying where the industry engineering gaps sit and where hardware-software co-design is the right framing.

,

Definition:

The 6G MIMO Research Agenda

A research agenda is a structured list of open problems paired with success criteria and rough time horizons. The 6G MIMO research agenda consolidated by ITU-R (2023), the Hexa-X project (EU, 2024), and the CommIT/Huawei 6G workshop (TU Berlin, 2023) identifies approximately thirty open questions across five themes:

  1. Channel modeling and measurement (spatial non-stationarity, near-field, multi-frequency consistency)
  2. Distributed and scalable architectures (cell-free, federated, edge-cloud split)
  3. Hardware-software co-design (full-duplex, holographic, low-resolution ADC)
  4. AI/ML integration (end-to-end training, learned receivers, data-driven channel models)
  5. New deployment scenarios (NTN, ISAC, RIS, ultra-dense indoor)

This section focuses on the problems introduced in Sections 27.1-27.5 plus the cross-cutting issues they surface.

,

Five Open Problems at a Glance

ProblemWhat is knownWhat is openPhD-scale projectTime horizon
Non-stationarity modelsMeasurement data from 5+5+ campaigns; theoretical frameworkNo tractable parametric model accepted by 3GPPPropose + validate a cluster-based visibility-region model and get it adopted in 3GPP Release 20+3-5 years
Ultra-dense cell-free scalabilityO(NAP3)\mathcal{O}(N_{\text{AP}}^3) barrier; consensus algorithms with fixed iterationsNo algorithm guarantees o(NAP2)o(N_{\text{AP}}^2) cost with centralized performanceDesign a message-passing algorithm with provable convergence on realistic AP graphs2-4 years
Full-duplex massive MIMOCancellation cascade up to ∼125\sim 125 dB; spatial nulling theoryNo commercial product within cost envelopeDemonstrate 100 MHz FD link with ≥1.8×\geq 1.8\times half-duplex SE at << 5 W additional power4-6 years
Holographic MIMOPizzo-Marzetta-Sanguinetti DoF theoremManufacturable surface realizing ≥60\geq 60 percent of theoretical DoFBuild a 30x30 cm, 8-bit phase, λ/6\lambda/6 spaced surface at 28 GHz and measure achievable DoF5-8 years
RIS + cell-free convergenceCascaded channel model; joint SDP formulationNo scalable joint optimization; no low-pilot channel estimationDevelop sparsity-exploiting cascaded channel estimator with provable sample complexity3-5 years

What Makes an Open Problem PhD-Scale?

A good PhD open problem has four properties:

  1. Tractable with existing tools. You can make progress without inventing a new branch of mathematics.
  2. Non-trivial. A determined undergraduate cannot solve it in a summer.
  3. Has a measurable success criterion. You can tell when you are done.
  4. Matters to someone. Industry, standardization bodies, or at least a well-defined research community cares.

Each open problem in the comparison table above satisfies all four. Each also admits smaller sub-problems that are reasonable first-year thesis projects and larger versions that would define a thesis. The right scale for an individual researcher is somewhere between "3 months to a workshop paper" and "3 years to a dissertation chapter."

Example: A First-Year Project in Each Area

For each of the five open problem areas, sketch a concrete first-year PhD project: a specific result to produce, the tools to use, and the deliverable venue.

Solution
Spatial non-stationarity

Project: Fit a two-cluster visibility-region model to the Lund and Eurecom measurement datasets (both are public). Tools: Expectation-maximization for cluster membership, Kolmogorov-Smirnov for goodness of fit. Deliverable: a publication at IEEE GLOBECOM with the fitted model and error comparison against WSS baseline. Sufficient for: a workshop paper + a journal submission.

Scalable cell-free

Project: Analyze the convergence rate of consensus MMSE on random geometric AP graphs (Poisson point process). Tools: Spectral graph theory, Grimmett's percolation bounds. Deliverable: an IEEE Transactions on Signal Processing submission relating the graph spectral gap to the SINR gap from centralized MMSE.

Full-duplex

Project: Measure the residual SI floor on an existing 8×88 \times 8 WARP or USRP testbed under varying PA backoff, characterizing the PA nonlinearity contribution. Tools: Hardware testbed, Wiener-Hammerstein model fitting. Deliverable: an Asilomar paper with measurement campaign results, feeding into a follow-up journal submission on joint cancellation and predistortion.

Holographic MIMO

Project: Simulate a λ/8\lambda/8-spaced 10×1010 \times 10 cm aperture at 2828 GHz using a full-wave solver and quantify the achievable DoF as a function of mutual-coupling correction. Tools: CST Studio Suite or HFSS, Slepian decomposition for DoF counting. Deliverable: an EuCAP conference paper with the gap between theoretical and numerical DoF.

RIS + cell-free

Project: Design a pilot protocol that recovers the cascaded channel for a ≤4\leq 4-RIS, ≤8\leq 8-AP toy deployment using O(Nt+NRIS)O(N_t + N_{\text{RIS}}) pilots via alternating minimization. Tools: Compressed sensing, sparse Bayesian learning. Deliverable: an ICC paper, then a TWC submission with convergence guarantees. ■\blacksquare

⚠️Engineering Note

Industry-Relevant Engineering Gaps

The research agenda above is academic. The complementary industry-engineering agenda — what a commercial vendor would pay consulting rates to see solved — is narrower and more concrete. At the 2023 CommIT/Huawei 6G workshop (TU Berlin), operator and vendor participants identified five concrete gaps that block deployment today:

  1. Calibration at scale: maintaining per-element RF calibration across 128+ elements under temperature and aging drift in an outdoor cabinet with ≤0.5\leq 0.5 percent air-interface overhead.
  2. Fronthaul cost: fiber-to-AP is the dominant deployment cost in cell-free networks; a fronthaul compression scheme that cuts it by 4×4\times is worth ∼30\sim 30 percent deployment savings.
  3. Power per bit: the E1 target for 6G is 10 pJ/bit; current best 5G mmWave radios sit at 500 pJ/bit. A 50×50\times efficiency gap.
  4. XL-MIMO beam management: beam-sweeping at >1000> 1000 narrow beams does not fit in the 5G reference signal budget; what is the right pilot structure?
  5. Mobility support: hand-off across ultra-dense cell-free neighborhoods at vehicular speeds has not been demonstrated beyond simulation.

Academic problems that touch one or more of these gaps are more likely to attract industry collaboration and funding.

Practical Constraints
  • •

    Power per bit: current 5G ∼500\sim 500 pJ/bit; target 6G ∼10\sim 10 pJ/bit

  • •

    Fronthaul cost dominates cell-free total capex at >50> 50 percent

  • •

    Beam management overhead: 5G NR allows ∼64\sim 64 SSBs per frame; proposals need compatible scaling

  • •

    Calibration drift tolerance: ±1\pm 1 dB amplitude, ±2∘\pm 2^\circ phase over 55 minutes

📋 Ref: ITU-R Report M.2516 (IMT-2030 Framework), 3GPP TR 38.855 (NR Positioning Study)
🎓CommIT Contribution(2023)

The 6G Workshop Synthesis

G. Caire, TU Berlin CommIT, Huawei 6G Research — TU Berlin / Huawei joint research workshop report, 2023

The 2023 joint CommIT / Huawei 6G research workshop brought academic and industry participants to TU Berlin for a four-day synthesis of the MIMO and AI/ML research frontier. The output — structured as fifteen open-problem statements — is the immediate inspiration for much of this chapter, including the industry-gap list above. The CommIT contribution to that workshop was the unification of the academic research agenda (treated in Sections 27.1-27.5) with the industry engineering gaps (treated in this section's engineering note).

The workshop also produced a consensus rank-ordering of the five research themes by ratio of "impact if solved" to "cost to solve." The top-ranked theme is distributed/scalable architectures (Section 27.2), followed by spatial non-stationarity modeling (Section 27.1). Holographic MIMO is ranked third, with full-duplex massive MIMO fourth because of its comparatively well-understood engineering barriers. The convergence of RIS and cell-free is ranked fifth but flagged for rapid promotion if channel estimation challenges are resolved.

6gresearch-agendacommithuaweiView Paper →

Why Hardware-Software Co-Design Is the Missing Piece

Four of the five open problems in this chapter are ultimately constrained by the interaction between the algorithm and the hardware it runs on: a PA non-linearity limits digital cancellation (Section 27.3); unit-cell spacing limits holographic DoF (Section 27.4); fronthaul bandwidth limits cell-free processing (Section 27.2); RIS phase resolution limits RIS-assisted beamforming (Section 27.5).

The traditional division of labor — hardware engineers design radios, signal processing engineers design algorithms — stops working when the fundamental limits of the algorithm are set by hardware parameters. The 6G research community has been slowly absorbing this lesson: the 2024 Hexa-X EU flagship explicitly asks for hardware-software co-design proposals, the IEEE SPAWC conferences have added hardware demo tracks, and graduate curricula (including this book) are being revised to integrate the two sides. The next decade's best MIMO research will likely come from researchers who can read a datasheet and a theorem with equal fluency.

,

Historical Note: Shannon's Warning About Closed Theories

1949-present

In his 1949 "Communication in the Presence of Noise" — the paper where he formalized the Shannon-Hartley theorem — Claude Shannon closed with a remark about the limits of his own theory: that information theory abstracts away the specific physical mechanism of communication, and that the engineering problem of getting close to capacity always requires re-engaging the physics. The warning has proved accurate at every generation boundary: 2G needed channel coding that Shannon had not explicitly described; 3G needed turbo codes and the iterative-decoding principle; 4G needed OFDM + MIMO + Turbo equalization; 5G added massive MIMO.

Every time, the information-theoretic envelope was known decades in advance; every time, closing the gap required engineering that the theory did not prescribe. The open problems of this chapter are the 6G instance of that pattern. The information theory is mostly worked out. The hardware and the algorithms to close the gap are not.

🔧Engineering Note

Closing Remark: What MIMO Research Looks Like After 2030

The research agenda in this section covers approximately the decade 2024-2034. What comes after is harder to predict — the next generation-boundary (8G? 7G?) will probably be as unexpected as the shift from macro cells to massive MIMO was in 2010. But some structural trends seem reliable:

  1. Physical layer research will continue to be driven by aperture-per-dollar scaling. Wherever apertures get cheaper, new MIMO architectures become feasible.
  2. Algorithm research will migrate toward learned, semi-learned, and data-assisted approaches, but the analytical baselines in this book will remain the performance bounds those algorithms are measured against.
  3. Standardization will continue to lag theory by 5-10 years, as it has for every prior generation. Research that wants to reach products needs to pick its battles early.
  4. Hardware-software co-design will become the norm, not the exception. The "pure theory" and "pure implementation" corners will remain valuable but the action will be in the middle.

The best MIMO research of the next decade will come from people who have both read this book and moved beyond it.

Practical Constraints
  • •

    Aperture cost (dollars per λ2\lambda^{2} of surface) is the first-order scaling parameter

  • •

    Standardization cycle: typically 3-4 years from first study item to frozen spec

  • •

    Academic-industry collaboration is the fastest path from open problem to deployment

,

Key Takeaway

The book is not closed. Each of the five open problems in this chapter has a concrete PhD-scale project attached to it, a clearly identified research community, and an industry engineering gap that rewards its solution. MIMO research in 2026 looks less like a mature field approaching its limits and more like a field whose best problems are in front of it. The theoretical framework developed in Chapters 1-26 gives the language for the research community to state these problems precisely; the work of answering them remains for the reader.

Why This Matters: Continuing Beyond This Book

The MIMO research agenda intersects the agendas of several other books in this library. Chapter 27 of the OTFS book addresses waveform design for high-mobility 6G. The RIS book closes on a parallel open-problems chapter covering near-field RIS and nonlinear active RIS. The Cell-Free book (when complete) will treat the ultra-dense processing problem in more depth. The Telecom book's final chapter addresses system-level deployment questions that bracket the per-link problems studied here. Researchers working on any of these problems should treat the library as a single document with many overlapping chapters, not a collection of independent books.

PhD-Scale Open Problem

A research question with a concrete success criterion, a clearly identified community that cares about the answer, tractability with existing tools, and sufficient depth to occupy a researcher for 2-4 years. The five open problems catalogued in Chapter 27 are intended to satisfy these criteria.

Related: Research Agenda, The 6G MIMO Research Agenda

Quick Check

According to the 2023 CommIT/Huawei 6G workshop, which research theme was ranked highest in the ratio of impact-if-solved to cost-to-solve?

Full-duplex massive MIMO

Holographic MIMO

Scalable distributed processing for ultra-dense cell-free

RIS + cell-free convergence

Correction:
Scalable distributed processing for ultra-dense cell-free

The workshop consensus placed distributed/scalable cell-free architectures first, with spatial non-stationarity modeling second. Holographic MIMO was third, full-duplex fourth, and RIS + cell-free fifth (with room to rise if channel estimation challenges are solved).

Convergence of RIS and Cell-Free ArchitecturesChapter Summary