Codebook-Based Effective CSI Feedback for Precoding Design in MIMO Systems

I. INTRODUCTION

• Physical-channel feedback can mismatch interference-aware combining, which depends on interference covariance and matters to coordinated transmission in multi-cell networks,,.Practical limits on reciprocity and the overhead of reporting full channel matrices also motivate explicit, structured CSI feedback –, –.Source: §I · pp. 1–2
• ECSI preserves the Type-II payload structure while directly compressing post-combining effective channels, eliminating the rank-truncation error component of conventional Type-II feedback.The framework supports iterative OTA precoder refinement and coordinated multi-cell precoding using feedback for interfering links.Source: §I · p. 2
• The paper derives a unified effective-channel estimation-error characterization that separates beam-selection, rank-truncation, and AWGN-induced components for conventional CSI and ECSI feedback.The analysis also identifies a structural advantage for ECSI in interference-limited regimes.Source: §I · p. 2

II. SYSTEM MODEL AND BEAMFORMING DESIGN

• The system model considers multi-cell downlink transmission with multiple BSs and UEs, linear precoders and combiners, and a sum-rate performance measure.The stated model assumes a narrowband flat-fading channel over a single subband.Source: §II · pp. 2–3

A. Beamforming Design Based on MSE

• Because the MSE is not jointly convex in the precoders and combiners, the design alternates between updating each set while holding the other fixed.The combiner update minimizes a stream’s MSE; the precoder update minimizes the sum of MSEs subject to the BS power constraint, with the resulting MMSE precoder derived from the KKT conditions.Source: A. Beamforming Design Based on MSE · p. 3

B. OTA Pilot Signaling

• With precoders fixed, BSs send orthogonal UE-specific CSI-RS or DMRS, which UEs use to estimate post-precoding effective channels and update their linear combiners.The pilots are associated with precoder streams and are orthogonal across streams and BSs.Source: B. OTA Pilot Signaling · p. 3
• Cell-specific beam pilots let UEs estimate serving and interfering BS beam-domain channels; the paper assumes a full unitary DFT codebook to simplify analysis.The channel estimates support CSI compression and feedback for precoder design; developments extend to reduced or structured beam sets.Source: B. OTA Pilot Signaling · pp. 3–4

III. BASELINE FRAMEWORK FOR CSI COMPRESSION AND PRECODING DESIGN

• Codebook-based reporting projects the channel onto a predefined beam domain and compresses it through rank reduction based on dominant eigenmodes.Temporal and frequency-domain compression can additionally exploit Doppler and sub-band correlation.Source: §III · p. 4
• The analysis focuses on a fixed time instant and a single subband, with further Doppler- and delay-domain compression stated to be incorporable under the same principles.Source: §III · p. 4
• The described procedure follows 3GPP NR CSI reporting, which is specified for the serving cell or, in coherent joint transmission, all active transmission points.Source: §III · p. 4

A. CSI Compression at the UE

• The UE estimates beam-domain channels, selects L beams with the largest per-beam energy, and forms a reduced beam-domain channel estimate.The beam set uses the corresponding selected columns of the codebook.Source: A. CSI Compression at the UE · p. 4
• The UE applies SVD and retains the R strongest singular values and their right singular vectors, where R ≤ min(N, L).This rank reduction further compresses the beam-domain channel before quantization and feedback.Source: A. CSI Compression at the UE · p. 4
• The UE quantizes compressed channel coefficients and feeds them back with beam indices; the BS uses these reports to reconstruct a reduced-dimension channel.The reconstruction preserves the dominant channel subspace up to a unitary transformation associated with an omitted component.Source: A. CSI Compression at the UE · p. 5

B. Precoding Design at the BS

• With serving-cell-only CSI, the design ignores inter-cell interference; with CSI from multiple BSs, it explicitly incorporates that interference into precoding.The converged precoder is used at the serving BS, while performance is evaluated using the UE’s true combiner and SINR.Source: B. Precoding Design at the BS · p. 5

IV. ECSI COMPRESSION AND PRECODING DESIGN

• Conventional reduced-dimension CSI omits UE combining, whereas ECSI feeds back post-combining effective channels that capture inter-stream, inter-UE, and inter-cell interference.The receive combiner implicitly reduces the receive dimension, replacing the baseline’s interference-agnostic rank reduction; OTA refinement can mitigate compression and quantization errors.Source: §IV · pp. 5–6

A. ECSI Compression at the UE

• For each stream, the UE combines the estimated channel and applies beam-domain compression; these operations are interchangeable.Using a common beam set across streams, the UE selects the L beams with highest received energy and forms the reduced-dimension effective channel.Source: A. ECSI Compression at the UE · p. 6
• The UE quantizes the compressed effective channel and reports it with normalization factors and beam indices for reconstruction at the BS.The reconstructed vectors capture the dominant effective-channel subspace for reduced-feedback precoding.Source: A. ECSI Compression at the UE · p. 6

B. Precoding Design at the BS

• Serving-cell-only ECSI neglects inter-cell interference, while ECSI from additional BSs enables its explicit suppression in precoder design.The additional feedback covers a subset of non-serving BSs; the formulation handles both availability scenarios.Source: B. Precoding Design at the BS · p. 6
• The BSs transmit UE-specific precoded pilots; UEs update their combiners, compute and report compressed ECSI, and the BSs reconstruct it for precoder updates.The procedure repeats after initialization with precoders from Algorithm 1.Source: B. Precoding Design at the BS · p. 6
• Refinement assumes the CSI-RS channel estimate remains valid across iterations, which requires them to span only a fraction of the channel coherence time.CSI-RS updates occur in an outer loop; the inner loop can be implemented centrally or distributively, with backhaul exchange when multiple BSs’ ECSI is available.Source: B. Precoding Design at the BS · p. 6

C. Effective Channel Error

• The comparison measures post-combining effective-channel error rather than physical-channel error because the effective channel determines precoder design.The joint average over channel realizations and UE combiners is intractable because the combiner depends on interference, precoding, and channel statistics.Source: C. Effective Channel Error · pp. 6–7
• For tractability, the error analysis fixes the channel realization and UE combiner and averages only over receiver AWGN.This isolates beam-selection, rank-truncation, and channel-estimation-noise effects.Source: C. Effective Channel Error · p. 7
• The AWGN-conditioned comparison is restricted to a single stream; sensitivity to the combiner choice is evaluated numerically.The stream index is omitted in the analysis.Source: C. Effective Channel Error · p. 7

1) Conventional CSI Feedback:

• For a fair comparison, the rank-reduced CSI uses a virtual combiner formed by projecting the true combiner onto the retained left-singular-vector basis.Applying that projected combiner to rank-reduced CSI gives the effective channel used in the comparison with the true antenna-domain effective channel.Source: 1) Conventional CSI Feedback · p. 7
• Beam-selection error is orthogonal to rank-truncation and AWGN errors because they occupy distinct beam subspaces.The decomposition separates discarded beams from errors associated with the selected-beam subspace.Source: 1) Conventional CSI Feedback · p. 7
• Noisy beam-energy estimates drive beam selection; the analysis replaces random TOP-L selection with a threshold chosen so the expected selected-beam count is L.The beam-selection error is evaluated using the resulting selection probabilities.Source: 1) Conventional CSI Feedback · p. 7
• The rank-truncation error depends on discarded singular modes and their alignment with the combiner, not only on the discarded singular values.The analysis uses a representative beam set rather than enumerating all selected subsets.Source: 1) Conventional CSI Feedback · p. 8
• At sufficiently high pilot SNR, a first-order subspace-perturbation approximation is used for the AWGN error, with higher-order cross terms neglected.Quantization error is neglected and normalization factors are assumed perfectly known to isolate the analyzed error sources.Source: 1) Conventional CSI Feedback · pp. 7–8
• The conventional-feedback effective-channel error is approximated as the sum of beam-selection, rank-truncation, and AWGN-induced terms.The rank-truncation term depends on both discarded singular values and the alignment of the fixed combiner with the corresponding discarded left singular vectors.Source: 1) Conventional CSI Feedback · p. 8

2) ECSI Feedback:

• Selecting beams by noisy metrics biases their reported coefficients, so the AWGN-induced error must be evaluated conditioned on the beam-selection event.The analysis expresses the conditional moment using Nuttall-Q functions and combines it with the beam-selection probabilities.Source: 2) ECSI Feedback · p. 9
• ECSI’s mean-squared effective-channel error consists of beam-selection and AWGN components, with no rank-truncation term.ECSI directly compresses the post-combining effective channel rather than applying an explicit rank-reduction step.Source: 2) ECSI Feedback · p. 9
• The beam-selection error reflects the loss from discarding non-selected beams; the analysis replaces random TOP-L selection with a threshold chosen to select L beams on average.At high pilot SNR, dominant-beam selection becomes effectively deterministic and the conditioning effect vanishes.Source: 2) ECSI Feedback · pp. 8–9
• Conventional CSI can reduce AWGN impact through rank-truncated projection, but adds rank-truncation error that can matter when the combiner overlaps the discarded subspace.This overlap can arise in interference-limited scenarios, where the combiner is shaped primarily by interference suppression.Source: 2) ECSI Feedback · p. 9
• ECSI extends to coordinated multi-TRP transmission by reporting post-combining effective CSI, avoiding the unitary-alignment mismatch of independently compressed local CSI.The mismatch can degrade coordinated precoding when local SVD subspaces are not aligned.Source: 2) ECSI Feedback · p. 9

D. Implementation and Feedback Overhead

• Iterative ECSI refinement uses standard downlink pilots multiplexed with data, avoiding additional downlink signaling overhead.Each update still requires a dedicated uplink ECSI report, even when the channel is static.Source: D. Implementation and Feedback Overhead · p. 10
• Each ECSI iteration requires L log2 M + S log2 qn + LS(log2 qm + log2 qp) feedback bits, including when the channel is quasi-static.The report includes L beam indices, quantized complex effective-channel coefficients for each stream, and S normalization coefficients.Source: D. Implementation and Feedback Overhead · p. 10
• When channel conditions change significantly, conventional CSI feedback can instead report a rank-R representation with overhead L log2 M + R log2 qn + LR(log2 qm + log2 qp).The ECSI mechanism reuses the payload structure of conventional type-II codebook-based reporting for backward compatibility.Source: D. Implementation and Feedback Overhead · p. 10

V. NUMERICAL RESULTS AND DISCUSSION

• 30% and 25% higher sum rate are achieved by ECSI-SC and ECSI-IC, respectively, than their conventional counterparts, with convergence within about five iterations.The coordinated schemes also outperform non-coordinated schemes by accounting for inter-cell interference.Source: Figure 1 · p. 10, §V · p. 11
• 42% and 48% higher 10th-percentile per-stream rates are achieved by CSI-IC and ECSI-IC, respectively, versus CSI-SC and ECSI-SC; ECSI gives up to 100% improvement over corresponding CSI schemes.The gains are most pronounced for cell-edge UEs.Source: Figure 2 · p. 10, §V · p. 11
• Approximately 2 dB lower effective-channel estimation error is achieved by ECSI than CSI when the UE combiner lies within the subspace retained by compressed CSI.In this setting, rank-truncation error for CSI-SC is zero, but CSI-selected beams are not necessarily optimal for representing the effective channel.Source: §V · p. 11, Figure 3 · p. 11
• Approximately 4 dB lower effective-channel estimation error is achieved by ECSI in the rank-truncation scenario, because it directly compresses the post-combining channel.The passage identifies this robustness as relevant when interference suppression leads the UE combiner to span a larger portion of the channel subspace.Source: §V · p. 11, Figure 3 · p. 11
• As the reported beam count increases, coordinated schemes gain over non-coordinated schemes because better serving- and interfering-channel representations improve interference suppression.At low beam counts, self-interference from channel estimation and compression errors outweighs inter-cell interference; at high counts, residual inter-cell interference limits non-coordinated schemes.Source: §V · p. 11, Figure 4 · p. 11
• Increasing BS antenna count improves all schemes, while the relative gain from inter-cell coordination becomes smaller as channels from different cells become more spatially separable.For a larger number of UEs, sum rate rises with added spatial streams, but interference also intensifies; ECSI-based schemes achieve larger gains than conventional CSI-only feedback, particularly in dense-user scenarios.Source: §V · pp. 11–12, Figure 6 · p. 12, Figure 5 · p. 12
• Higher magnitude and phase quantization resolution increases sum rate; coordination gains emerge as quantization error falls and interference suppression improves.At low quantization resolution, self-interference from quantization errors dominates, so coordinated schemes do not outperform non-coordinated counterparts.Source: §V · p. 12, Figure 7 · p. 12
• With feedback beam count fixed at L = 8, increasing angular spread raises compression error and reduces sum rate across schemes.Greater angular spread requires more angular components to represent the channel, while increasing overlap between UE and cell angular supports further reduces achievable sum rate.Source: §V · p. 12, Figure 8 · p. 12

• ECSI achieves up to 30% sum-rate improvement over conventional CSI-based precoding, with particularly strong gains for cell-edge and interference-limited UEs.Source: §VI · p. 13
• UEs report compact post-combining effective channels, accounting for UE combining and eliminating rank-truncation mismatch in conventional physical-channel feedback.Source: §VI · p. 13
• The framework reuses the existing codebook feedback structure and supports OTA iterative precoder refinement using standard downlink pilots without additional downlink signaling overhead.Source: §VI · p. 13
• Analytical error characterizations distinguish conventional CSI and ECSI feedback by beam selection, rank truncation, and AWGN-induced errors.Source: §VI · p. 13