⚡ electrical engineering

Structured Code-Domain Matrix Calibration for Shared-Chain Digital Beamforming Arrays

This paper proposes a regularized code-domain matrix estimation and compensation framework that effectively mitigates RF-induced crosstalk in shared-chain digital beamforming arrays, significantly improving beamforming, nulling, and angle-estimation performance compared to unstructured least-squares approaches.

Original authors: Xinhua Dai, Kai Xie, Youkang Wang

Published 2026-07-14

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Xinhua Dai, Kai Xie, Youkang Wang

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you have a super-smart radio team with 32 antennas, but to save money and space, you've only given them 4 microphones. That's a lot of antennas for just a few mics! To make this work, the team uses a clever trick: they assign each antenna a unique "secret code" (like a specific rhythm or pattern). All the antennas shout their messages at once, mixed together into one big signal that goes into the 4 microphones. Later, a computer uses those secret codes to untangle the mess and hear what each antenna was saying individually. This is called Shared-Chain Digital Beamforming.

But here's the glitch: in the real world, nothing is perfect. The microphones have a tiny bit of "memory" (they hold onto a sound for a split second too long), the timing is slightly off, and the filters used to untangle the codes aren't perfect. Because of this, when the computer tries to listen to Antenna #1, it accidentally hears a faint echo of Antenna #2, #3, and #4. It's like trying to listen to a friend at a party, but their voice is constantly bleeding into your ear from three other conversations happening nearby. This "bleeding" is called code-domain leakage.

The Problem with Old Solutions

For years, engineers tried to fix this by just adjusting the volume and timing of each antenna individually. They thought, "If I just turn down the volume on Antenna #2, the echo will go away." But the paper shows that this old way is like trying to fix a leaky roof by only painting the ceiling. It misses the real problem: the mixing of the signals. The authors argue that simply fixing the volume (gain) and timing (phase) isn't enough because the "bleeding" between channels is a complex, structured mess that depends on the specific codes being used.

The New "Leakage Fingerprint" Solution

The authors propose a new way to look at the problem. Instead of just looking at the volume of each antenna, they treat the whole system like a giant, slightly messy mixing board. They realized that the "bleeding" isn't random; it follows a specific pattern based on the codes used.

They developed a special math tool (a regularized structured estimator) that acts like a detective. Instead of guessing, it looks at a short "calibration" signal (a known test tone) and maps out exactly how much each antenna is leaking into the others. It creates a "leakage fingerprint."

Think of it this way: If the old method was just turning a single knob for each antenna, this new method draws a detailed map of every single wire connecting the microphones. It knows exactly which wires are crossed and how much signal is jumping the gap.

What the Simulations Showed

The authors didn't just guess; they ran thousands of computer simulations to test their idea. Here is what they found:

The Old Way vs. The New Way: When they compared their new method to the old "volume-only" fix, the new method reduced the error by about 18 dB. That's a huge difference—like turning down the static on a radio so much you can finally hear the music clearly.
Beating the "Generic" Fix: They also compared it to a "brute force" method that tries to fix the whole mess without using the code patterns. Their new method was still about 6 dB better. This proves that using the specific "secret code" patterns to guide the math makes a real difference.
Getting Close to Perfect: In their simulations, their method recovered almost all the performance of a "perfect" system (what they call "oracle" performance).
- Beamforming (Finding the Target): Without any fix, the system struggled to find a target. With their fix, the signal quality improved by about 1.45 dB compared to the old method.
- Blocking Noise (Nulling): The system got much better at silencing interfering noises, improving "null suppression" to about 48 dB.
- Pinpointing Direction: The error in guessing the direction of a signal dropped from about 0.68° (without fixing) to just 0.06° (with their fix). That's incredibly precise!

A Bonus Trick: Swapping the Codes

The paper also suggests a fun extra step. If the system detects that the current "secret codes" are causing too much leakage (maybe due to a hardware glitch), it can swap in a different set of codes from a library. Crucially, this doesn't require changing any hardware or adding more microphones; it just changes the digital rhythm. The simulations showed that picking the "healthiest" code set could squeeze out a little extra performance, like finding a better seat at the concert to hear the music better.

The Bottom Line

The paper concludes that for these compact, code-sharing radio systems, you can't just fix the volume of each antenna. You have to understand and fix the specific "crosstalk" caused by the codes themselves. By mapping this "leakage fingerprint" and using it to clean up the signal, the system can perform almost as well as if it had a perfect microphone for every single antenna, all while using far fewer hardware parts. It's a smart, software-based fix for a hardware limitation.

Technical Summary: Structured Code-Domain Matrix Calibration for Shared-Chain Digital Beamforming Arrays

Problem Statement
Shared-chain Spread Spectrum Digital Beamforming (SSDBF) offers a hardware-compressed alternative to fully digital arrays by multiplexing multiple aperture channels through a common analog RF/IF path. While this reduces hardware scale, it fundamentally alters the observation model for downstream signal processing. In conventional arrays, calibration typically addresses element-wise gain-phase errors or aperture-domain mutual coupling. However, in shared-chain SSDBF, finite RF memory, timing skew, switching transitions, and demultiplexing filters couple the nominally orthogonal code channels. This results in a "code-domain leakage" where recovered virtual channels are contaminated by residual crosstalk from other physical channels. Conventional calibration methods fail to address this specific structured linear mixing, leading to degraded beamforming, nulling, and angle-estimation performance.

Methodology
The paper formulates the impairment as a structured code-domain matrix estimation and compensation problem. The core methodology involves the following steps:

Signal Modeling: The effective receiver response is modeled as $z[n] = A_C x[n] + v[n]$ , where $A_C$ is an equivalent code-domain calibration matrix. This matrix is decomposed into a diagonal gain-phase component ( $g$ ) and an off-diagonal leakage fingerprint matrix ( $L_C$ ). The off-diagonal entries represent the SSDBF-specific crosstalk induced by the shared chain and the deployed codebook.
Structured Estimation: A regularized estimator is developed to estimate $A_C$ from short calibration frames. Unlike unstructured least squares (LS), this estimator exploits "product-code spectral priors." It applies weighted shrinkage to off-diagonal leakage entries based on the spectral location of the product codes relative to the demultiplexing filter passband. Entries associated with product codes far from the passband are shrunk more aggressively.
Noise Handling: The framework supports optional colored-noise whitening if the post-demultiplexing noise covariance is estimated from empty calibration slots, improving statistical efficiency.
Compensation and Metrics: The estimated matrix is used to compute low-dimensional receiver-state metrics (leakage energy, matrix conditioning, and channel drift). These metrics guide a regularized inverse compensation applied to working snapshots before adaptive beamforming.
Optional Codebook Update: If a library of candidate codebooks with the same cardinality is available, the estimated leakage structure can drive a task-aware update of the active codebook to minimize a composite cost function (balancing leakage, conditioning, and task utility) without altering the physical hardware topology.

Key Contributions
The paper makes four primary contributions:

Observation Model: Development of a shared-chain code-domain observation model that explicitly describes how finite RF memory, timing skew, and demultiplexing filters induce off-diagonal leakage.
Decomposition: Introduction of a leakage-fingerprint matrix decomposition to separate ordinary diagonal gain-phase errors from SSDBF-specific code-domain crosstalk.
Structured Estimator: Formulation of a regularized structured estimator that incorporates product-code spectral priors and optional noise whitening, specifically tailored for finite-snapshot calibration.
Theoretical and Empirical Validation: Derivation of identifiability conditions and a baseline finite-snapshot error law under orthogonal pilots, verified through behavioral RF-aware simulations.

Results
Behavioral RF-aware simulations with finite calibration snapshots demonstrate the efficacy of the proposed method:

Calibration Accuracy: The proposed structured estimator reduces the matrix estimation error by approximately 6 dB relative to unstructured least squares and by 18 dB relative to diagonal-only calibration. Specifically, the leakage-fingerprint error reaches approximately -20 dB, compared to -13 dB for unstructured LS.
Task Performance: The compensation recovers most of the "oracle" (ideal matrix) performance.
- Output SINR: Improves from -1.2 dB (no calibration) to 1.2 dB (proposed), approaching the ideal 1.45 dB.
- Null Suppression: Increases from ~41 dB (no calibration) to ~48 dB (proposed), nearing the ideal 49 dB.
- Angle Estimation: Reduces error from 0.68° to 0.06°, close to the ideal 0.03°.
Codebook Update: The optional same-cardinality codebook update step yields modest but consistent improvements in SINR (~~0.2 dB), null suppression (~~0.8 dB), and angle error (~0.02°) by selecting a codebook that minimizes diagnostic cost under fixed hardware constraints.

Significance and Claims
The paper claims that code-domain leakage estimation serves as an effective signal processing layer specifically for shared-chain digital beamforming arrays. It argues that while conventional mutual-coupling and array-shape calibration methods address aperture-domain response errors, they do not directly estimate the internal code-domain leakage matrix experienced by the SSDBF processing chain. By treating the problem as a structured matrix estimation rather than a simple gain-phase calibration, the proposed method enables the recovery of high-fidelity beamforming and angle-estimation performance in hardware-compressed architectures. The authors position this work as a necessary calibration step for fixed-topology shared-chain systems, noting that it can be combined with conventional aperture calibration in a layered stack. Future work is identified as integrating this layer with aperture mutual-coupling calibration and temperature-dependent drift modeling.