Abhay Kumar Sah

Reduced Neighborhood Search Algorithms for Low Complexity Detection in MIMO Systems

Neighborhood search algorithms such as likelihood ascent search (LAS) and reactive tabu search (RTS) have been proposed for low complexity detection in multiple-input multiple-output (MIMO) systems having a large number of antennas. Both these algorithms are iterative and search for the vector which minimizes the maximum likelihood (ML) cost in the neighborhood. In this paper we propose a way to reduce the size of the neighborhood. For this, we propose a metric and a selection rule to decide whether or not to include a vector in the neighborhood. We use the indices of, say

An MMP-Based Approach for Detection in Large MIMO Systems Using Sphere Decoding

K

An Unconstrained Likelihood Ascent Based Detection Algorithm for Large MIMO Systems

, largest components of the metric for generating a reduced neighborhood set. This reduced set is used to evaluate the performance of the resulting LAS and RTS algorithms. Simulation results show that this reduces the complexity significantly while maintaining the error performance. We also show that the proposed reduced neighborhood algorithms can make MIMO systems with several hundred antenna pairs feasible.

Stopping Rule-Based Iterative Tree Search for Low-Complexity Detection in MIMO Systems

Sphere decoders (SDs) are known to provide near maximum-likelihood performance in multiple-input multiple-output systems. But their ability to provide good error performance has not evinced any interest in large antenna systems because of their very high complexity. However, if the number of erroneous dimensions can be made low, say by some pre-processing, we can use SD for those dimensions. This could potentially improve the error performance at a reasonable complexity. To identify such dimensions, we present a multipath matching pursuit-based approach. Simulation results show that compared to existing algorithms, the proposed approach can provide a significant improvement in error performance.

Beyond fixed neighborhood search in the likelihood ascent algorithm for MIMO systems

Neighborhood search algorithms have been proposed for detection in large multiple-input multiple-output systems. They iteratively search for the best vector in a fixed neighborhood. A better way could be to look for an update which is not restricted to a fixed neighborhood. Motivated by this, we formulate a problem to maximize the reduction in maximum likelihood (ML) cost and use it to derive an expression for updating the current solution. Using this update and a likelihood function regarding the locations of errors, we propose an unconstrained likelihood ascent search (ULAS) algorithm. ULAS seeks to provide the maximum reduction in ML cost by finding an update which is not restricted to be in a fixed neighborhood. Using simulations, the proposed algorithm has been shown to provide better error performance for uncoded systems than existing algorithms, at lower complexity. We also show that ULAS is amenable to lattice reduction, which helps in obtaining two variants leading to further improvements in performance.

Quasi-Orthogonal Combining for Reducing RF Chains in Massive MIMO Systems

Breadth first tree search (BFTS) algorithms are known to provide a close to maximum likelihood (quasi-ML) solution at a low-complexity if the received sequence is detected in the right sequence order. However, finding the right sequence order has an exponential overhead. In view of this, we propose to repeatedly apply a BFTS algorithm to all sequence orders. Since it will test all the orders, it is expected to achieve quasi-ML performance. However, this will increase the complexity because of redundant iterations. The complexity can be reduced if we can stop the iterations as soon as a quasi-ML solution is achieved. For this, we propose two stopping rules, one relies on a constellation based heuristic and the other one uses the distribution of ML cost. It is found that their complexity curves have a cross-over point. Thus, a combination of the two rules provides a quasi-ML error performance at a low-complexity for uncoded as well as coded systems. We further show that the proposed stopping rule can reduce the complexity of depth first tree search algorithms also. Last, for large MIMO systems, compared with existing algorithms, it is found to be exceptionally better in terms of both error performance and complexity.

Improved Sparsity Behaviour and Error Localization in Detectors for Large MIMO Systems

Neighborhood search algorithms have been proposed for detection in large/massive multiple-input multiple-output (MIMO) systems. They iteratively search for the best vector in a fixed neighborhood. However, the ML solution may not lie in the searched space or the search may take a large number of intermediate vectors to converge. Instead of searching in a fixed neighborhood, a better way will be to look for an update which is not restricted to be in a fixed neighborhood. Motivated by this, we formulate an optimization problem to maximize the reduction in ML cost and use it to derive an expression for updating the solution. We use a metric based on the channel matrix and the error vector to determine the likelihood of a symbol being in error. Using this likelihood and the update, we propose a likelihood ascent search (LAS) algorithm to find an update which is not restricted to be in a fixed neighborhood and seeks to provide maximum reduction in ML cost. This process continues till there is a reduction in the ML cost. Compared to existing LAS based algorithms, it is found to provide better error performance, that too at a lower complexity.

Improving the reliability of receivers for 5G networks

We address the issue of reducing the number of radio frequency (RF) chains in large/massive multiple input multiple output systems while retaining their detection performance. For this, we use a hybrid combining approach and formulate a signal to interference plus noise ratio (SINR) maximization problem. We argue that using an orthogonal basis of the channel matrix to maximize the numerator of the SINR is a worthwhile approach to pursue and use it to propose a quasi-orthogonal combining scheme. Simulation results show that the proposed scheme can provide a detection performance and sum rate close to that of a system utilizing a dedicated RF chain for each receive antenna with less number of RF chains and without increasing the overall computational complexity.

Likelihood-Based Tree Search for Low Complexity Detection in Large MIMO Systems

Sparsity based techniques have been found to be promising for detection in large/massive multiple-input multiple-output (MIMO) systems. They initialize with an initial solution vector, localize its erroneous symbols and then correct them. However, the process works only if the errors in the initial solution vector are sparse enough and are localized accurately. In this paper, we improve the localizing capability by proposing a modification in the residual update computed by the generalized orthogonal matching pursuit algorithm. Subsequently, we show that the sparsity behaviour can be improved by concatenating error vectors to create a larger vector, with no increase the complexity. Combining both these strategies, we propose to concatenate MMSE solution vectors (say

Iterative matrix inversion based low complexity detection in large/massive MIMO systems

K