---
title: Inter-Band Beam Configuration (IBBC)
url: https://www.emergentmind.com/topics/inter-band-beam-configuration-ibbc
type: topic
---

# Inter-Band Beam Configuration (IBBC)

Searching arXiv for the cited IBBC-related papers and closely related terminology.
arxiv_search(query="all:(\"Inter-Band Beam Configuration\" OR \"beam squinting compensation\" OR \"joint band assignment and beam management\" OR \"coordinated beam selection out-of-band\" OR \"dual-band reconfigurable intelligent surface\" OR \"sub-THz radio unit selection\" )", max_results=10, sort_by="submittedDate")

Inter-Band Beam Configuration (IBBC) denotes a family of cross-frequency beam-control mechanisms in which beam parameters, beam states, or beam geometry in one band or subband are used to configure, compensate, infer, or coordinate beam operation in another. In the literature represented here, the term appears in several technically distinct but related forms: subband beam-squint compensation for network-controlled repeaters (NCRs), hierarchical control for joint band assignment and beam management, out-of-band (OOB) beam selection from sub-6 GHz to mmWave, shared-aperture dual-band reconfigurable intelligent surface (RIS) operation, and user-equipment (UE) orientation modeling through the angle between low-band and high-band broadside directions [2402.10368], [2308.13202], [1903.12589], [2406.02975], [2507.09244]. Taken together, these usages indicate that IBBC is not a single algorithm but a cross-band design paradigm for maintaining beam consistency, reducing training overhead, and exploiting structural relations between bands.

## 1. Terminological scope and formal definitions

In a subband beamforming setting, IBBC is realized by taking each beamformer designed at the “measurement” band \(f_c\), calculating a simple per-antenna phase-shift vector that exactly cancels the extra delay \(\Delta f \cdot R_n \cdot U^*/c\), and then applying these corrected weights whenever transmitting in band \(f_c+\Delta f\) [2402.10368]. In that usage, the central problem is beam squinting: signaling related to measurements is transmitted in one subband centered at frequency \(f_c\), while data transmission is performed at a different frequency \(f_c+\Delta f\), so a frequency-dependent array radiation pattern can lead to beam misalignment.

In a dual-band UE formulation, Inter-Band Beam Configuration is defined as the angle between the broadside vectors of the low-band array and the high-band array. With \(\mathbf{u}_{\mathrm{LB}}\) and \(\mathbf{u}_{\mathrm{HB}}\) denoting the unit-vectors pointing along the broadside of the low-band and high-band arrays, respectively, the IBBC angle is written as
\[
\theta_{\mathrm{IBBC}}=\arccos\!\bigl(\mathbf{u}_{\mathrm{LB}}^{T}\mathbf{u}_{\mathrm{HB}}\bigr).
\]
When the high-band beam is one of a finite steering grid, \(\beta\) can be two-dimensional, written as
\[
\beta=[\,\theta_{\mathrm{IBBC}}^{\mathrm{az}},\;\theta_{\mathrm{IBBC}}^{\mathrm{el}}\,]^T
\]
[2507.09244]. In that setting, IBBC captures the UE’s orientation and the relative beamforming direction at low- versus high-band.

In OOB-aided multi-user mmWave MIMO, the same cross-band idea appears as an inter-band mapping from sub-6 GHz beams to mmWave candidate beams. Because sub-6 GHz beams are wider, one sub-6 GHz beam pair \((n,m)\) “covers” several mmWave beam pairs through a 3 dB mapping set \(S(n,m)\) [1903.12589]. In hierarchical reinforcement learning (HRL) for multi-band communication, IBBC is cast as a sequential decision problem in which a high-level policy selects the band and a low-level policy sets beam-management thresholds in the selected band [2308.13202]. In dual-band RIS design, IBBC refers to independently reconfigurable beam steering in sub-6 GHz and mmWave within a single aperture [2406.02975].

These formulations are not identical. A plausible implication is that IBBC should be understood as an umbrella concept covering cross-band beam coupling, rather than a uniquely standardized protocol primitive.

## 2. Array-theoretic basis and subband beam-squint compensation

The most explicit signal model appears in the NCR-assisted subband framework. A planar antenna array of \(N\) elements is located at positions \(R_n \in \mathbb{R}^3\), \(n=1,\ldots,N\). In many examples, a Uniform Linear Array (ULA) along \(y\) is used:
\[
R_n=[0,(n-(N+1)/2)\cdot d,0]^T,\qquad d=\lambda_c/2,\qquad \lambda_c=c/f_c.
\]
A plane wave departing toward azimuth \(\theta\) and optionally zenith \(\phi\) at frequency \(f\) induces the element-wise phase
\[
a_n(\theta,f)=\exp\!\bigl(-j2\pi f/c \cdot R_n \cdot U(\theta,\phi)\bigr),
\]
with
\[
U(\theta,\phi)=[\sin\phi\cos\theta,\;\sin\phi\sin\theta,\;\cos\phi]^T,
\]
and steering vector
\[
a(\theta,f)=[a_1(\theta,f),\ldots,a_N(\theta,f)]^T.
\]
Given complex weights \(w=[w_1,\ldots,w_N]^T\), the array factor is
\[
AF(\theta,f)=w^H a(\theta,f)=\sum_{n=1}^N w_n e^{-j2\pi f/c\,R_n\cdot U(\theta,\phi)}.
\]
The wide band is split into two or more subbands, with \(f_1=f_c\) for measurement and \(f_2=f_c+\Delta f\) for data [2402.10368].

If the beamformer \(w\) is designed to point at true direction \(\theta_0\) at \(f_1\), then at \(f_2\) the peak shifts to \(\theta_0+\Delta\theta\), where for an ideal ULA
\[
\Delta\theta=-\tan(\theta_0)\cdot(\Delta f/f_1).
\]
For \(\Delta f \ll f_1\), and for small \(\theta_0\) in radians, this becomes
\[
\Delta\theta \approx -\theta_0\cdot(\Delta f/f_1),
\]
with stated validity for \(|\Delta f/f_c|<0.05\) and \(|\theta_0|<60^\circ\) [2402.10368].

The compensation mechanism uses the phase-deviation view
\[
AF(\theta,f_2)=\sum_n w_n e^{-j2\pi f_1/c\,R_n\cdot U}\cdot e^{-j2\pi \Delta f/c\,R_n\cdot U},
\]
where the extra per-element phase
\[
d_n(\theta)=\exp(-j2\pi \Delta f/c\,R_n\cdot U)
\]
produces the squint. Let \(U^*=(\theta_0,\phi_0)\) be the intended beam direction found by maximizing \(AF\) at \(f_1\). Define the compensation vector
\[
c_n=d_n^*(U^*)=\exp(+j2\pi \Delta f/c\,R_n\cdot U^*).
\]
The compensated weights at \(f_2\) are then
\[
\tilde w_n(f_2)=w_n(f_1)\cdot c_n,\qquad n=1,\ldots,N,
\]
or, in vector form,
\[
w(f_2)=w(f_1)\odot [\exp(j2\pi \Delta f/c\,R_n\cdot U^*)]_{n=1\ldots N}.
\]
This closed-form update is directly realizable in analog or hybrid beamforming because only phase shifts are applied, and it also supports a codebook approach in which a compensated version of each nominal beam is pre-stored for each subband [2402.10368].

## 3. Control, performance metrics, and implementation constraints

In the NCR setting, downlink performance is measured through
\[
\mathrm{SINR}_k(f)=\frac{|h_k(f)^H w_k(f)|^2}{\sigma^2+\sum_{j\neq k}|h_k(f)^H w_j(f)|^2},
\]
where \(h_k(f)\) includes path loss, shadowing and small-scale fading. The instantaneous rate is
\[
R_k(f)=B\cdot \log_2(1+\mathrm{SINR}_k(f)),
\]
or it is mapped to discrete MCS based on a 10% BLER target [2402.10368].

The reported numerical result is specific: in a 3GPP-style system-level simulation at 28 GHz with a ULA of 256 elements, without compensation and \(\Delta f=1\,\mathrm{GHz}\), the 10%-ile user’s SINR fell by \(\approx 20\,\mathrm{dB}\) versus the \(\Delta f=0\) reference. With compensation, the SINR CDF nearly collapses onto the \(\Delta f=0\) curve, recovering the \(20\,\mathrm{dB}\) loss and restoring 10%-ile throughput [2402.10368]. This supports the narrower claim that subband IBBC can make measurement-band beam decisions usable for data-band transmission without material performance loss when the compensation model is available.

The operational cost is also tightly specified. Per subband, the computational burden is one complex-vector multiplication of length \(N\), or codebook corrections can be pre-computed off-line. The only side information needed is \(U^*\), the beam’s pointing direction, already known from beam management. No new pilot signals or feedback are required; the method uses existing beam-report procedures such as SSB/CSI-RS \(\rightarrow\) RSRP \(\rightarrow\) best-beam index. Control-plane impact is described as negligible: IBBC requires tagging each subband’s beam index or weight set in the scheduling grant, and the NCR control link simply forwards the compensated beam index [2402.10368].

For multiple subbands, compensation generalizes by computing
\[
c_n^{(k)}=\exp(+j2\pi \Delta f_k/c\,R_n\cdot U^*)
\]
for subband \(k\) at \(f_k=f_c+\Delta f_k\), then forming
\[
w^{(k)}=w(f_1)\odot c^{(k)}.
\]
For multi-beam or multi-sector operation, the same process is repeated for each beam direction \(U_m^*\) [2402.10368].

Several limitations are explicit. Endfire ambiguity arises when codebook entries produce symmetric beams and user-side direction feedback is required to disambiguate. For large \(\Delta f\) or elevated \(\phi\) tilt, compensation is exact only for the main lobe; side lobes may still shift. Finite-bit phase shifters introduce quantization error in \(c_n\). Under fast dynamic mobility, if \(U^*\) changes faster than the time required to update \(c^{(k)}\), residual squint may degrade performance [2402.10368].

## 4. Cross-band beam selection and hierarchical decision frameworks

Cross-band IBBC is also used to reduce beam-training overhead by transferring information from a low-frequency band to a high-frequency band. In coordinated OOB beam selection, the sub-6 GHz channel \(\hat H^u\) is projected into a spatial spectrum
\[
S^u=W^H \hat H^u V \in \mathbb{C}^{M_{BS}\times M_{UE}},
\]
and the side-information is the expectation \(E[|S^u|^2]\), where the \((m,n)\)-th entry gives the average sub-6 GHz gain seen by UE \(u\) with transmit beam \(n\) and receive beam \(m\) [1903.12589]. No extra training overhead is required because this expectation is obtained from standard sub-6 GHz CSI measurements.

A central mapping is the 3 dB relation between coarse sub-6 GHz beams and candidate mmWave beams. For each sub-6 beam pair \((n,m)\), the set
\[
S(n,m)=S_{UE}(n)\times S_{BS}(m)
\]
contains all mmWave beam pairs whose 3 dB lobes overlap the chosen sub-6 GHz beams [1903.12589]. The fine mmWave search is then restricted to that candidate set. In the uncoordinated version, each UE chooses \((n_u^{un},m_u^{un})\) to maximize its own average single-user SNR. In the hierarchical coordinated version, UEs are ordered, and each UE overhears via low-rate device-to-device links the chosen sub-6 receive beam indices of lower-ranked UEs, then solves a constrained optimization to avoid receive-beam conflicts [1903.12589].

The key analytical statement is Proposition 1: in the large-array limit, after ZF combining at the base station, the average SINR of user \(u\) is
\[
E[\gamma^u(\vec n,\vec m)] =
\begin{cases}
g_{n_u,m_u}/\sigma^2_{\tilde n}, & \text{if } m_u\neq m_w\ \forall\, w\neq u \\
0, & \text{if } \exists\, w\neq u:\ m_w=m_u
\end{cases}
\]
[1903.12589]. The practical interpretation in the source is that any two UEs sharing the same mmWave BS beam cause zero effective SINR for one of them under ZF, described there as irreducible co-beam interference. The coordination rule therefore forbids candidates whose receive beam conflicts with already selected beams.

The reported performance summary is again explicit. For \(K=5\) UEs, \(N_{BS}=64\), \(N_{UE}=16\), mmWave at 28 GHz, and sub-6 GHz at 3 GHz, coordinated IBBC yields up to \(20\)–\(30\%\) sum-rate gain over uncoordinated selection at moderate SNR \(0\)–\(5\) dB and small inter-UE distances \(<15\) m. The D2D exchange requires only \(\log_2(M_{BS})\) bits per UE at beam-update intervals, and beam coherence times are stated to be much larger than channel coherence times [1903.12589].

A different cross-band control view is developed in the HRL formulation for joint band assignment and beam management. Time is indexed by \(m=1,\ldots,M\), each \(m\) being one OFDM time frame. The state
\[
s_m=\{\hat F_m,\hat W_m,\hat H_m,\{\hat H[k,m]\}_k,\{H[k,m]\}_k,\{\underline H[k,m]\}_k\}
\]
captures currently installed analog and digital beams and codebook indices in each band, together with recent spectral-efficiency feedbacks [2308.13202]. The high-level action or goal is
\[
g_m\in\{0,1\},
\]
indicating the selected band, with \(0=\) sub-6 GHz and \(1=\) mmWave. The low-level action is
\[
a_m=(\tau_m,\bar\tau_m),
\]
two continuous thresholds in the chosen band. Reward is built from the instantaneous spectral efficiency
\[
R_m=(1-b_m)(1/\underline N)\sum_{k=1}^{\underline N}\underline S[k,m]+b_m(1/K)\sum_{k=1}^{K}S[k,m],
\]
and
\[
r_m=c(a_m)\cdot R_m,
\]
where \(c(a_m)=0\) during beam training and \(c(a_m)=1\) when data is sent [2308.13202].

The hierarchy separates a slow-timescale upper-level policy \(\pi^H(s_n^H)\to g_n\) from a fast-timescale lower-level policy \(\pi^L(s_m,g_n)\to a_m\). Both levels use DDPG with experience replay and slow-moving target networks, and off-policy correction is done either by direct importance sampling or action-relabeling [2308.13202]. The evaluation uses a Manhattan grid at \(40\,\mathrm{km/h}\), vehicle density \(10\,\mathrm{veh/km}\), QuaDRiGa “3GPP UMi” ray-tracing, \(32\) Tx/\(16\) Rx mmWave hybrid arrays with \(8\) RF chains and \(4\) streams, and \(4\times 4\) fully digital sub-6 GHz arrays with \(4\) streams. The metric is ensemble-averaged cumulative data rate over \(1\,000\) channel realizations. At \(30\,\mathrm{dBm}\), HRL achieves \(\approx 27\,\mathrm{Mbps}\) versus \(\approx 6.5\,\mathrm{Mbps}\) for DRL and \(\approx 4\,\mathrm{Mbps}\) for a greedy mmWave-only policy, and HRL converges in \(\approx 20\) episodes while DRL takes \(\approx 60\) episodes for \(\approx 6.5\,\mathrm{Mbps}\) [2308.13202].

These two strands—OOB beam transfer and hierarchical band/beam control—address different problems, but both instantiate IBBC as a mechanism for exploiting inter-band structure to reduce the effective cost of beam management.

## 5. IBBC as a latent geometry variable in RU selection

In sub-THz radio unit selection, IBBC is elevated from a beam-control primitive to a latent geometry variable. The end-to-end mapping for a UE at unknown 3-D position \(i\) is written as
\[
b_i=f(\zeta_i,\Omega_i,\beta_i),
\]
where \(\zeta_i\in\mathbb{R}^{N_\zeta}\) is a sub-10 GHz channel feature vector, \(\Omega_i\) is the UE’s 3-D orientation, \(\beta_i\) is the IBBC, and \(b_i\in\{1,2,\dots,K\}\) is the best sub-THz RU index [2507.09244]. If the network ignores \(\Omega_i\) and \(\beta_i\), the learned function \(\tilde f(\zeta_i)\) suffers from “one-to-many” ambiguity.

The framework uses three supervised deep-learning classifiers. With \(\mathbf{h}_i\in\mathbb{R}^{N_h}\) as a measured sub-THz feature vector from the coarsely presumed-best RU, the inputs are
\[
\mathbf{x}_i^{(1)}=
\begin{bmatrix}
\zeta_i\\
h_i
\end{bmatrix},
\qquad
\mathbf{x}_i^{(2)}=
\begin{bmatrix}
\zeta_i\\
\beta_i
\end{bmatrix},
\qquad
\mathbf{x}_i^{(3)}=\zeta_i,
\]
corresponding respectively to IBBC inference, RU inference with IBBC, and RU inference without IBBC [2507.09244]. The IBBC label is quantized into one of \(16\) classes, and the RU index is chosen among \(K=9\) RUs. All three classifiers share a fully connected architecture with three hidden layers of sizes \([128,50,8]\), ReLU hidden activations, a Softmax output, categorical cross-entropy loss, SGDM with learning rate \(0.01\), batch size typically \(128\), and an \(80\%/20\%\) training-validation split [2507.09244].

The inference loop is staged. Algorithm 3 first produces a coarse RU estimate without IBBC. The network then measures \(\mathbf{h}_i\) on that RU, forms \(\mathbf{x}_i^{(1)}\), applies Algorithm 1 to infer \(\hat\beta_i\), and finally refines the RU estimate by running Algorithm 2 on \([\zeta_i;\hat\beta_i]\) [2507.09244]. This procedure is explicitly motivated by the observation that IBBC indicates beamforming information or UE orientation, which is typically not shared with the network as part of signalling.

The simulation uses an \(8\times 5\times 3\) m office room, \(9\) ceiling-mounted RUs, one wall-mounted sub-10 GHz AP, UEs on a \(0.5\) m grid, one of \(16\) IBBCs assigned at random, \(2\times 2\) URAs at UE and RU/AP, carriers at \(5.8\) GHz and \(100\) GHz, and ray-traced instantaneous power-delay profiles as features [2507.09244]. The reported SNR CDFs show median UE SNR \(\simeq 15\) dB for RU inference without IBBC, \(\simeq 22\) dB for perfect IBBC, and \(\simeq 20\) dB for inferred IBBC. At the \(20\)th percentile, inferred IBBC improves by \(4\)–\(5\) dB over the no-IBBC baseline. Final classification accuracies are \(\simeq 60\%\) for RU inference without IBBC, \(\simeq 75\%\) with true IBBC, and \(\simeq 50\%\) for the \(16\)-class IBBC inference task [2507.09244].

This usage differs conceptually from subband compensation. Here IBBC is not a phase-correction vector but a hidden state variable linking low-band channel fingerprints to high-band beam or RU choice.

## 6. Hardware embodiment in dual-band RISs

A hardware realization of independent inter-band beam configuration appears in a shared-aperture dual-band sub-6 GHz and mmWave RIS [2406.02975]. The mmWave RIS element at 28 GHz is a double-layer microstrip patch aperture-fed through a ground slot and loaded with a 1-bit reflection-type phase shifter using an MA4GP907 PIN diode. The two states are “off” with \(\phi\approx 0^\circ\) and “on” with \(\phi\approx 180^\circ\), and the measured \(|S_{11}|\) is approximately \(-2\) dB to \(-1\) dB. The reflection coefficient is
\[
\Gamma_{\mathrm{PS}}=
\begin{cases}
A_0 e^{j0^\circ}, & \mathrm{PD\ off}\\
A_1 e^{j180^\circ}, & \mathrm{PD\ on}.
\end{cases}
\]

The sub-6 GHz RIS element at 3.5 GHz reuses an \(8\times 8\) block of 28 GHz patches. The central \(6\times 6\) patches are selectively interconnected with \(3\) PIN-diode RF switches, giving \(2^3=8\) reconfigurable states. A multi-port model describes the internal connections through an adjacency matrix \(Y\), a switching vector \(x\), and an inter-port loading matrix \(Z_L(x)\). The scattered field per state is
\[
E_s(\Omega)=\sum_{m=1}^{M} i_m E_m(\Omega)+E_{\mathrm{oc}}(\Omega),\qquad
\mathbf{i}=-(\mathbf{Z}+\mathbf{Z}_L(x))^{-1}\mathbf{v}_{\mathrm{oc}}
\]
[2406.02975].

Beam steering is based on the continuous phase law
\[
\varphi_{mn}(f)=-k(f)\bigl(x_m\sin\varphi\cos\theta+y_n\cos\varphi\cos\theta\bigr),\qquad
k(f)=\frac{2\pi f}{c}.
\]
At 28 GHz, the desired phase is quantized to \(\{0^\circ,180^\circ\}\). At 3.5 GHz, each of the \(8\) states yields a distinct reflection phase, and a phase-entropy metric
\[
H=-\sum_{i=1}^{8} p_i\log_2 p_i,\qquad
p_i=\Delta\varphi_i/360^\circ,\qquad
\sum_i \Delta\varphi_i=360^\circ
\]
is used to ensure uniform spacing of the states [2406.02975].

Independent operation is achieved through physical and control isolation. A planar spiral inductor (PSI) is optimized so that at 28 GHz it is near-invisible to the patch, with \(\Delta S_{11}<0.2\) dB, while at 3.5 GHz it acts as a low-impedance metal link. A suspended electromagnetic band gap (EBG) structure is used to suppress surface waves, with a simulated bandgap of approximately 27–29 GHz. The FPGA writes sub-6 and mmWave registers over separate SPI buses or multiplexes, and the source states that no cross-coupling was measured: mmWave beam patterns remained unchanged over all 8 sub-6 states [2406.02975].

The experimental ranges are also explicit. The fabricated sub-6 GHz RIS with \(4\times 4\) sub-6 elements achieves beam steering from \(-35^\circ\) to \(35^\circ\). The \(8\times 8\) mmWave RIS achieves beam steering from \(-30^\circ\) to \(30^\circ\). At 28 GHz, reported gain is about \(6.5\) dBi with side-lobe level at most \(-10\) dB; at 3.5 GHz, efficiency is about \(60\%\) with circa \(4\) dBi broadside gain [2406.02975]. This hardware instantiation shows that IBBC can be realized not only algorithmically but also as an independently reconfigurable electromagnetic structure.

## 7. Limitations, misconceptions, and synthesis across the literature

A recurring misconception would be to treat IBBC as a single universally accepted mathematical object. The cited works do not support that simplification. In one line of work, IBBC is a per-element phase compensation across subbands [2402.10368]. In another, it is an HRL decomposition of band selection and beam management [2308.13202]. In OOB multi-user MIMO, it is a cross-band beam-candidate restriction and coordination mechanism [1903.12589]. In dual-band RIS design, it is independent beam steering in two bands on a shared aperture [2406.02975]. In sub-THz RU selection, it is the angle between low-band and high-band broadside vectors, or a two-dimensional azimuth-elevation quantity capturing UE orientation [2507.09244].

The limitations are equally heterogeneous. Subband compensation is exact only for the main lobe under large \(\Delta f\) or elevated \(\phi\) tilt, can be affected by finite-bit phase shifter quantization, and can degrade under fast mobility if \(U^*\) changes too quickly [2402.10368]. Coordinated OOB selection depends on exchange of beam-related information over low-rate D2D links and on the large-array approximation underlying the zero-SINR co-beam result [1903.12589]. HRL performance depends on the choice of goal horizon \(H\), off-policy correction strategy, and beam-training overhead model [2308.13202]. The dual-band RIS relies on PSI resonance, EBG tuning, and separate bias/control networks to preserve electrical decoupling [2406.02975]. RU selection with inferred IBBC remains limited by imperfect IBBC classification, with final accuracy around \(50\%\) on the \(16\)-class task [2507.09244].

Across these variants, a consistent technical pattern nevertheless emerges. Beam behavior in one band or subband is treated as informative for another band, but only after accounting for geometric mismatch, hardware constraints, training overhead, or latent orientation variables. This suggests that the core significance of IBBC lies in exploiting cross-band structure without assuming that beams are naively transferable. In that narrower and evidence-supported sense, IBBC functions as a unifying design principle for multi-band and multi-subband beam management in mmWave, sub-THz, and dual-band RIS systems [2402.10368], [2308.13202], [1903.12589], [2406.02975], [2507.09244].

Source: https://www.emergentmind.com/topics/inter-band-beam-configuration-ibbc