[Paper Review] A Random-List Based LAS Algorithm for Near-Optimal Detection in Large-Scale Uplink Multiuser MIMO Systems

Alexandre A. Jr Pereira, Raimundo Sampaio‐Neto|arXiv (Cornell University)|Dec 4, 2014

Advanced Wireless Communication TechniquesEngineering14 references16 citations

TL;DR

This paper proposes a random-list based likelihood-ascent search (RLB-LAS) detector for large-scale uplink MIMO systems, using iteratively generated random starting points from matched filter output to achieve near-maximum-likelihood (near-ML) performance with only O(K²) complexity per symbol. The algorithm outperforms existing LAS and MIMO detectors in both bit error rate (BER) and computational efficiency, approaching single-antenna AWGN performance even at K=N=20.

ABSTRACT

Massive Multiple-input Multiple-output (MIMO) systems offer exciting opportunities due to their high spectral efficiencies capabilities. On the other hand, one major issue in these scenarios is the high-complexity detectors of such systems. In this work, we present a low-complexity, near maximum-likelihood (ML) performance achieving detector for the uplink in large MIMO systems with tens to hundreds of antennas at the base station (BS) and similar number of uplink users. The proposed algorithm is derived from the likelihood-ascent search (LAS) algorithm and it is shown to achieve near ML performance as well as to possess excellent complexity attribute. The presented algorithm, termed as random-list based LAS (RLB-LAS), employs several iterative LAS search procedures whose starting-points are in a list generated by random changes in the matched filter detected vector and chooses the best LAS result. Also, a stop criterion was proposed in order to maintain the algorithm's complexity at low levels. Near-ML performance detection is demonstrated by means of Monte Carlo simulations and it is shown that this performance is achieved with complexity of just O(K^2) per symbol, where K denotes the number of single-antenna uplink users.

Motivation & Objective

• Address the high-complexity detection problem in large-scale uplink MIMO systems with tens to hundreds of antennas.
• Develop a low-complexity detector that approaches maximum-likelihood (ML) performance without requiring matrix inversion.
• Reduce computational load compared to traditional ML, sphere decoding, or MMSE-based detectors in massive MIMO scenarios.
• Improve convergence speed and detection accuracy over existing LAS variants like MIV-LAS and MSCS-LAS.
• Achieve near-ML performance with minimal increase in complexity as system size scales.

Proposed method

• The algorithm uses the matched filter (MF) output as the initial detection vector.
• It generates multiple random starting-point vectors by perturbing the MF result in each iteration.
• Each iteration performs a one-stage likelihood-ascent search (LAS) that greedily selects symbol changes reducing the ML cost function.
• The algorithm employs a stop criterion based on the ML cost of the best result so far and the theoretical minimum cost of an error-free decision.
• The final decision is the LAS result with the lowest ML cost among all iterations.
• The search uses a greedy ordering strategy that selects the symbol change providing the maximum reduction in ML cost at each step.

Experimental results

Research questions

• RQ1 Can a randomized iterative LAS approach achieve near-ML performance in large-scale uplink MIMO with significantly reduced complexity?
• RQ2 How does the performance of RLB-LAS compare to established detectors like MMSE-MLAS, MMSE-SIC, and MIV-LAS in terms of BER and complexity?
• RQ3 Does using random perturbations of the matched filter output as starting points improve detection performance compared to fixed or deterministic initial vectors?
• RQ4 What is the asymptotic complexity scaling of the proposed RLB-LAS algorithm with respect to the number of users K?
• RQ5 Can the proposed stop criterion effectively limit computational load while preserving near-ML performance?

Key findings

• The MF-RLB-LAS detector achieves a BER of 10⁻³ at only 1 dB below the AWGN-only SISO channel performance when K=N=20.
• For K=N=20, the average number of floating-point operations per symbol is 3×10⁴, significantly lower than the 1.1×10¹² required by ML detection.
• The complexity of MF-RLB-LAS scales as O(K²), while ML complexity grows exponentially with K, making it feasible for large-scale systems.
• The proposed algorithm outperforms MMSE-MLAS (3-stage), MMSE-SIC, and MMSE-SIC-MB in both BER and computational complexity.
• The performance of MF-RLB-LAS is nearly identical to that of the more complex MMSE-RLB-LAS variant, confirming the effectiveness of using MF as the base vector.
• The stop criterion successfully limits iterations and maintains low complexity without degrading performance, even as K increases.