Environment-Aware Hybrid Beamforming by Leveraging Channel Knowledge Map

I. INTRODUCTION

• Hybrid beamforming needs accurate CSI despite limited RF-chain observations, while estimating its large, intertwined channel coefficients requires substantially more training.Compressive-sensing methods also require complex observation-matrix design and iterative processing, with substantial training still needed for high observation gain.Source: §I · pp. 2–3
• Beam sweeping searches transmit–receive beam pairs sequentially, and with 64 BS and 16 UE antennas can take up to 5.2 seconds even for analog beamforming alone.The search overhead grows with antenna count or codebook size.Source: §I · p. 3
• This work extends prior CKM-enabled, training-free analog beam alignment from single-RF-chain, single-stream links to general hybrid analog/digital beamforming.It leverages UE location information and proposes CAM and BIM as two CKM types for the hybrid design.Source: §I · pp. 3–4
• The proposed codebook-based hybrid architecture selects analog beams from predetermined codebooks and freely designs digital beams; perfect-CSI optimization provides a rate upper bound but requires prohibitive training and computation.The optimal analog-beam selection and digital-beamforming design assumes perfect CSI.Source: §I · p. 4
• CAM predicts location-specific path AoAs and AoDs for light-training channel reconstruction, whereas BIM supplies candidate beams and enables direct beamforming design from a reduced-dimension effective channel.CAM still requires computation to select optimal beams after estimating the MIMO channel; BIM avoids that prohibitive computation.Source: §I · p. 4
• The CKM-enabled schemes significantly improve effective communication rate over CS-based and location-based benchmarks, even with moderate UE location errors, by saving real-time training overhead.Simulations use communication environments generated with Wireless Insite.Source: §I · p. 4

• Section V presents extensive numerical results using commercial ray-tracing channels, showing advantages over benchmark methods in effective communication rate and training overhead.Source: §I · p. 5
• The paper draws its conclusion in Section VI.Source: §I · p. 5

II. SYSTEM MODEL

• The transmitter and receiver use codebook-based analog precoders and combiners, with phase-shifter entries constrained to constant modulus.The analog beamforming columns are selected from predetermined codebooks.Source: §II · p. 6
• Once the analog matrices are fixed, the baseband precoder and combiner are freely designed for the lower-dimensional effective channel.The signal model first applies analog beamforming and then digital baseband processing.Source: §II · p. 6
• Channel coefficients remain constant during each coherent block of N symbol durations and may vary between blocks.The channel matrix for block t is denoted H[t].Source: §II · p. 6
• After noise whitening, the effective baseband channel is an AWGN MIMO channel whose capacity-achieving input is CSCG signaling.For sufficiently long blocks, the system’s average spectral efficiency is expressed in bits per second per Hertz.Source: §II · p. 7

III. OPTIMAL HYBRID BEAMFORMING WITH PERFECT CSI

• For each analog precoder and combiner choice, the optimal baseband solution follows from the effective channel’s SVD and water-filling power allocation.The selected analog matrices are then chosen to maximize average spectral efficiency.Source: §III · pp. 8–9
• Finding the globally optimal analog beamformers requires exhaustive search over all transmitter and receiver codebook choices.The resulting optimal algorithm is described as suitable only for upper-bounding performance in small-size MIMO systems because of its high computational cost.Source: §III · p. 9
• Estimating the full channel matrix requires at least MrMt measurements per coherent block, which is prohibitive when MrMt is much larger than the number of RF chains.Compressed-sensing approaches can reduce training overhead but rely on additional assumptions such as spatial sparsity and known array geometry.Source: §III · p. 9
• For large MIMO systems, channel-training-based beamforming has a huge performance gap from the maximum rate Ropt with perfect CSI.The passage identifies this gap when the antenna-dimension product is comparable to the training requirement.Source: §III · pp. 10–11
• The paper proposes CKM-enabled environment-aware hybrid beamforming to address the training and computation burdens using user location and site-specific channel information.The CKM is described as a database of location-specific channel-related information.Source: §III · p. 11

A. Key Idea of CKM-Enabled Environment-Aware Communications

• CKM captures the site-specific influence of the propagation environment so channel knowledge can be obtained using the UE’s location.The paper motivates this approach because accurate mathematical channel models are difficult to obtain in complex environments.Source: A. Key Idea of CKM-Enabled… · p. 12
• A channel matrix map (CMM) maps each potential UE location directly to an estimated channel matrix, but requires excessive storage.The CMM may vary with changes in the surrounding propagation environment, distinct from changes caused by UE movement.Source: A. Key Idea of CKM-Enabled… · pp. 12–13
• Because propagation environments usually change more slowly than UE locations, CKM updates are generally needed only after significant environmental change.The passage notes that this timescale relationship also applies when environmental factors such as pedestrians may vary at a similar timescale to UE locations.Source: A. Key Idea of CKM-Enabled… · pp. 12–13
• A channel path map (CPM) stores location-specific path information to reconstruct the channel with reduced storage, but adapting it to environmental change requires updating the complete map.Its stored path information includes quantities such as significant-path power, phase, and angles of arrival and departure.Source: A. Key Idea of CKM-Enabled… · p. 13
• The proposed channel angle map (CAM) and beam index map (BIM) target dynamic environments while requiring only light real-time training.They are introduced to overcome the CPM’s difficulty adapting to environmental changes.Source: A. Key Idea of CKM-Enabled… · pp. 13–14

B. CAM-Enabled Hybrid Beamforming

• The full discretized angle grid can still impose substantial training and computation costs, and without angle priors CS observation design may require many measurements.The passage states the grid resolution should exceed antenna size, yet the resulting search remains costly even with CS.Source: B. CAM-Enabled Hybrid Beamforming · p. 15
• CAM maps UE location to candidate AoA/AoD tuples, whose smaller search space can significantly reduce real-time channel-training overhead.CAM focuses on path angles while leaving path gains available for real-time training to accommodate environment variations.Source: B. CAM-Enabled Hybrid Beamforming · p. 15
• When CAM includes the ground-truth angles, the known array responses let light training estimate path gains and reconstruct the channel.If CAM includes extra candidate angles, their gains are represented as zeros, so training estimates gains over the candidate set.Source: B. CAM-Enabled Hybrid Beamforming · pp. 15–16
• CAM can also support existing CS estimation by designing observations around location-specific angles and reducing OMP measurements and computational complexity when the candidate set is much smaller.Without angle prior information, CS uses the complete angle grid for observation-matrix design; CAM supplies location-specific angles to better direct training power.Source: B. CAM-Enabled Hybrid Beamforming · p. 18
• Even when CAM misses ground-truth angles, gain estimation projects unpredicted-path gains onto predicted angles, allowing their channel effects to be partially reconstructed.The estimated gains contain contributions from both included paths and projections of unpredicted-path gains.Source: B. CAM-Enabled Hybrid Beamforming · p. 18
• Increasing the number of paths predicted by CAM can improve prediction accuracy.Source: B. CAM-Enabled Hybrid Beamforming · p. 18
• Using UE location and CAM angle information, the method reconstructs the channel after real-time pilot training estimates gains only for candidate paths.The candidate paths are selected based on the UE location.Source: B. CAM-Enabled Hybrid Beamforming · pp. 18–19
• Location-specific candidate angles can enable more effective precoding and combining matrices, potentially increasing SNR during training.The training phase uses the candidate angle subset from CAM.Source: B. CAM-Enabled Hybrid Beamforming · p. 19
• CAM-enabled design significantly reduces the performance gap with environment-unaware training-based hybrid beamforming, reducing unknown parameters from M_rM_t to L_hat[t].The passage attributes the reduction to the smaller number of unknown parameters.Source: B. CAM-Enabled Hybrid Beamforming · p. 19

C. BIM Enabled Hybrid Beamforming

• BIM maps each UE location to candidate transmit and receive analog beams, avoiding CAM’s intermediate channel reconstruction.The selected candidates form reduced codebook subsets from which the analog beamforming matrices are determined.Source: C. BIM Enabled Hybrid Beamforming · p. 19
• Light beam sweeping measures candidate beam-pair powers, and the analog beam selection maximizes their aggregate channel power.The direct optimization requires the channel matrix and has prohibitive computational complexity; the proposed sweeping method avoids that requirement.Source: C. BIM Enabled Hybrid Beamforming · p. 20
• BIM reduces the beam pairs to be trained from |F||W| to |F̂||Ŵ|, cutting the number of required training durations.The paper states that the candidate subsets are substantially smaller than the complete codebooks.Source: C. BIM Enabled Hybrid Beamforming · pp. 22–23
• BIM reduces computational complexity because only one matrix SVD is computed after the analog beams are selected.The selected analog matrices define an equivalent channel used to obtain the digital beamforming matrices.Source: C. BIM Enabled Hybrid Beamforming · pp. 22–23
• Compared with CAM, BIM simplifies implementation and storage, while CAM provides propagation-channel information independent of array configuration and codebook.The choice between the maps depends on the application scenario.Source: C. BIM Enabled Hybrid Beamforming · p. 23

D. Construction of CAM and BIM

• CAM/BIM performance depends on UE location accuracy and the quality of map construction.The paper notes that sufficiently accurate location acquisition is becoming more feasible and gives LiDAR and RTK as examples of centimeter-level localization technologies.Source: D. Construction of CAM and BIM · p. 23
• Map data can come from online channel-training results or offline methods such as ray tracing, using physical-environment information to reconstruct site-specific channels.The paper also mentions specialized measurement equipment and combining online and offline collection to obtain sufficient data.Source: D. Construction of CAM and BIM · pp. 23–24
• After collecting finite data points, KNN, Kriging-based methods, or DNNs can learn the CKM; construction cost is independent of real-time communication and does not occupy communication time.The paper also cites an EM-based CKM construction method in [32].Source: D. Construction of CAM and BIM · p. 24

V. SIMULATION RESULTS

• BIM-enabled schemes drastically reduce training slots from more than 1000 to less than 50 for Mt = 256; proposed CAM- and BIM-enabled schemes also significantly reduce required training slots versus OMP.Source: §V · p. 26
• CAM and BIM effective rates increase with BS antenna count and significantly outperform OMP, while LS-based rates decrease as training overhead grows.The paper attributes the LS decline to training overhead outweighing beamforming gain in large antenna systems.Source: §V · p. 26, Figure 9 · p. 27
• With 3 m average location error, both proposed schemes still significantly outperform benchmark schemes, as the error only slightly degrades beamforming gain.The paper states that the training-overhead savings dominate because overhead affects communication rate linearly while beamforming gain affects it logarithmically.Source: Figure 9 · p. 27, §V · p. 27

VI. CONCLUSION

• The proposed CKM-enabled environment-aware hybrid beamforming technique aims to reduce prohibitive training overhead while achieving high rate performance.Source: §VI · pp. 27–28
• The paper introduces two CKM types, CAM and BIM, and presents hybrid beamforming schemes enabled by them.Source: §VI · p. 28
• In ray-tracing simulations over a practical propagation environment, the proposed schemes significantly outperform benchmarks even with moderate UE-location errors.Source: §VI · p. 28

PROOF OF LEMMA1

• Under vec(˜N) ∼ CN(0, σ^2 I_M²s), the MSE in (28) can be written in the stated form.Source: PROOF OF LEMMA1 · p. 28
• One bound step follows because the arithmetic mean is larger than the harmonic mean.Source: PROOF OF LEMMA1 · p. 28
• Another step uses 1/λ_l((Q̂ᴴQ̂)⁻¹) = λ_l(Q̂ᴴQ̂).Source: PROOF OF LEMMA1 · p. 28
• The derivation also invokes the Khatri-Rao product Gram identity and the Cauchy-Schwarz inequality.Source: PROOF OF LEMMA1 · p. 28