Abstract
This paper presents a novel optimum combiner designed to enhance the time-averaged signal-to-interference-plus-noise ratio (SINR) for maximal-ratio combining (MRC) in environments characterized by spatially correlated Nakagami-$m$ fading. By integrating Principal Component Analysis (PCA) with eigenvalue decomposition (EVD), the proposed system decorrelates received signals and projects them into an orthonormal basis. Leveraging the "theorem of irrelevance," a Principal Component Selection Combining (PCSC) technique is introduced to filter out dimensions with low SINR, thereby mitigating noise enhancement and reducing computational complexity. Simulation results demonstrate significant SINR improvements and a substantial reduction in signal processing overhead.
Introduction: The Challenge of Spatial Correlation
In modern wireless communication, diversity techniques like Maximal-Ratio Combining (MRC) are essential for mitigating fading. However, as mobile devices shrink and antenna density increases, the assumption of independent fading across antenna branches often fails. This leads to spatial correlation, where signals and noise become "colored" or interdependent across branches. Standard MRC receivers, which typically assume white noise and independent signals, suffer performance degradation in these scenarios. Furthermore, practical constraints like imperfect channel estimation introduce additional noise, further complicating signal recovery.
Proposed Architecture: PCA-Based Decorrelation
To address these issues, the authors propose a receiver architecture that incorporates a spatial decorrelator before the combining stage. The process begins with an analytical derivation of the covariance matrix for the Nakagami-$m$ fading channel. Using Eigenvalue Decomposition (EVD), the system identifies the principal components of the correlated signal space.
As shown in the architecture above, the decorrelator projects the received signals into a new space spanned by orthonormal basis functions. This transformation effectively "whitens" the spatial domain, allowing the subsequent stages to treat the components as independent.
Principal Component Selection Combining (PCSC)
The core innovation lies in the application of the "theorem of irrelevance." The authors argue that not all dimensions in the transformed signal space contribute equally to the final signal quality. By implementing a PCSC matrix that functions as an ideal spatial low-pass filter, the receiver retains only the dominant components—typically those containing over 90% of the total signal power.
Discarding the "irrelevant subspace" (dimensions with low eigenvalues) prevents the inclusion of components that are dominated by noise and interference. This "addition by subtraction" approach not only raises the overall SINR but also prevents the noise enhancement that occurs when MRC is applied to correlated noise.
Performance and Results
The proposed method was evaluated using a novel scattering simulator based on 3GPP LTE specifications. The results highlight a clear advantage over traditional Maximum Likelihood (ML) estimation methods.
In an eight-antenna configuration, the PCSC method achieved a time-averaged SINR of 6.2906, compared to 4.8994 for standard ML estimation—a gain of approximately 3.08 dB. The PDF shift in the figure above visually confirms that the proposed combiner consistently delivers higher signal quality.
Beyond signal quality, the method offers significant efficiency gains. As spatial correlation increases, the number of components required for processing drops to between 25% and 50% of the total antenna branches. This reduction in dimensionality translates directly into lower computational power requirements for the receiver.
Conclusion
The proposed optimum combiner effectively mitigates the negative impacts of spatial correlation in Nakagami-$m$ fading channels. By combining PCA-based decorrelation with selective component combining, the system achieves higher SINR, reduced fading duration, and a dramatic decrease in processing complexity. This approach provides a robust framework for future high-density antenna systems, such as Massive MIMO, where spatial correlation is an inherent challenge.
Summary updated