Abstract
As a green MIMO structure, the heterogeneous hybrid analog-digital ($H^2AD$) MIMO architecture has been shown to own a great potential to replace the massive or extremely large-scale fully-digital MIMO in the future wireless networks to address the three challenging problems faced by the latter: high energy consumption, high circuit cost, and high complexity. However, how to intelligently sense the number and direction of multi-emitters via such a structure is still an open hard problem. To address this, we propose a two-stage sensing framework that jointly estimates the number and direction values of multiple targets. Specifically, three target number sensing methods are designed: an improved eigen-domain clustering (EDC) framework, an enhanced deep neural network (DNN) based on five key statistical features, and an improved one-dimensional convolutional neural network (1D-CNN) utilizing full eigenvalues. Subsequently, a low-complexity and high-accuracy DOA estimation is achieved via the introduced online micro-clustering (OMC-DOA) method. Furthermore, we derive the Cramér-Rao lower bound (CRLB) for the $H^2AD$ under multiple-source conditions as a theoretical performance benchmark. Simulation results show that the developed three methods achieve 100% number of targets sensing at moderate-to-high SNRs, while the improved 1D-CNN exhibits superior under extremely-low SNR conditions. The introduced OMC-DOA outperforms existing clustering and fusion-based DOA methods in multi-source environments.
Introduction
The evolution towards 6G and advanced radar systems relies heavily on massive MIMO technology. However, fully digital massive MIMO architectures face significant challenges regarding high energy consumption, circuit costs, and computational complexity. Heterogeneous Hybrid Analog-Digital ($H^2AD$) MIMO has emerged as a "green" alternative. While $H^2AD$ architectures mitigate phase ambiguity through subarray heterogeneity, jointly sensing the number of emitters and their directions remains a complex problem. Traditional Direction of Arrival (DOA) algorithms often assume a known number of targets, which is impractical in dynamic environments like UAV tracking. This paper addresses the "open hard problem" of intelligently sensing both the count and the coordinates of multiple emitters simultaneously.
Proposed Two-Stage Sensing Framework
The researchers propose a two-stage sensing framework designed for $H^2AD$ MIMO receivers. The first stage focuses on target number sensing (counting), while the second stage performs high-accuracy DOA estimation.
Stage 1: Target Number Sensing
Three methods are developed to infer the number of emitters from the eigenvalues of the received signal's covariance matrix:
- Improved Eigen-Domain Clustering (EDC): An unsupervised method using the DBSCAN algorithm to separate "signal" clusters from "noise" clusters in a transformed eigenvalue space.
- Enhanced DNN: A supervised network utilizing five engineered statistical features, such as eigenvalue entropy and standard deviation, to classify the number of sources.
- Improved 1D-CNN: A deep learning approach that processes raw, log-transformed eigenvalue sequences to learn signal patterns directly.
Stage 2: DOA Estimation
Once the target count is established, the Online Micro-Clustering (OMC-DOA) method is employed. It uses the ESPRIT algorithm to generate candidate directions and a real-time clustering mechanism to filter pseudo-solutions, ensuring high precision with significantly lower computational overhead than traditional exhaustive searches.
Deep Learning for Target Counting
The 1D-CNN approach is particularly noteworthy for its performance in extreme conditions. By applying a logarithmic transformation to the eigenvalues, the network can more effectively identify "energy transitions" indicative of signal sources. This transformation compresses the range of eigenvalues, which can vary by orders of magnitude at low SNR.
Results and Discussion
Extensive Monte Carlo simulations were conducted using a massive array of 97 antennas partitioned into heterogeneous groups. Key findings include:
- Low-SNR Dominance: The 1D-CNN maintains approximately 80% accuracy even at a punishing SNR of -20 dB, where other methods begin to fail.
- Perfect Sensing: All three proposed counting methods achieve 100% accuracy once the SNR reaches -14 dB.
- Efficiency and Precision: The OMC-DOA method is orders of magnitude faster than exhaustive search methods like WGMD. Furthermore, in medium-to-high SNR scenarios, the DOA estimation accuracy successfully approaches the Cramér-Rao Lower Bound (CRLB), performing near the physical limits of the hardware.
Conclusion
The proposed $H^2AD$-based framework provides a robust and energy-efficient solution for joint target sensing in next-generation wireless networks. By combining subarray heterogeneity with advanced deep learning techniques, the system overcomes traditional limitations like phase ambiguity and the requirement for a known target count. Future work will explore wideband signal complexities and the impact of real-world hardware impairments.
Summary updated