Warm-Start Quantum Approximate Optimization Algorithm for QAM MIMO Data Detection

1. Problem Statement

The paper addresses the challenge of data detection in large-scale Multiple-Input Multiple-Output (MIMO) systems utilizing high-order Quadrature Amplitude Modulation (QAM) (e.g., 16-QAM, 64-QAM).

• The Core Difficulty: The optimal solution for MIMO detection is the Maximum Likelihood (ML) detector. However, its computational complexity grows exponentially with the number of transmit antennas and the constellation size ($O(|S|^{N_t})$), rendering it impractical for large-scale systems.
• The Modulation Challenge: Modern wireless standards use Gray coding to minimize bit error rates. This introduces a nonlinear mapping between symbols and bits. Consequently, the detection problem cannot be easily formulated as a standard Quadratic Unconstrained Binary Optimization (QUBO) problem (which works for BPSK/QPSK). Instead, it transforms into a Higher-Order Unconstrained Binary Optimization (HUBO) problem involving complex multi-variable interactions and a highly non-convex optimization landscape.
• The Quantum Gap: While Quantum Approximate Optimization Algorithm (QAOA) is a promising hybrid quantum-classical approach, existing applications to MIMO detection have largely been limited to simplified BPSK schemes. Extending QAOA to Gray-coded M-QAM is difficult due to the HUBO structure, and standard QAOA suffers from poor convergence, sensitivity to initialization, and performance degradation on Noisy Intermediate-Scale Quantum (NISQ) hardware.

2. Methodology

The authors propose a hybrid quantum-classical framework called WSLR-QAOA (Warm-Start Linear-Ramp QAOA). The methodology consists of four key components:

A. HUBO Formulation via Gray Coding

• The ML detection objective is first relaxed into a Semidefinite Programming (SDP) problem.
• The problem is then mapped to a quantum Hamiltonian. Unlike standard QUBO approaches, the authors derive a HUBO Hamiltonian that directly incorporates the nonlinear Gray-coded symbol-to-bit mapping.
• For 16-QAM, the real-valued symbol $s_k$ is expressed as a polynomial of binary variables, which are then mapped to Pauli-Z operators ($Z$). This results in a cost Hamiltonian ($H_C$) containing 2-body, 3-body, and 4-body interaction terms.

B. Warm-Start Initialization (BM-BCD)

Instead of initializing the quantum state in a uniform superposition (standard QAOA), the proposed method uses a problem-informed warm start:

1. Classical Relaxation: The SDP relaxation of the MIMO problem is solved using a Burer–Monteiro Block Coordinate Descent (BM-BCD) algorithm. This is chosen over interior-point methods (like CVX) for its superior scalability ($O(N^2K)$ vs. $O(N^{6.5})$).
2. Soft Estimates: The BM-BCD solution provides continuous "soft" estimates of the bits. These are converted into probability distributions for each qubit.
3. State Preparation: The quantum circuit is initialized in a product state where each qubit is rotated to match these probabilities (using $R_Y$ gates), rather than a uniform $|+\rangle$ state. This biases the search toward promising regions of the solution space.

C. Problem-Informed Mixer

• Standard QAOA uses a transverse-field mixer ($H_M = \sum X_i$).
• The proposed method introduces a Warm-Start Mixer ($H_{WS}$) constructed from the soft probabilities obtained in the initialization step. This mixer is tailored to the specific problem instance, facilitating better exploration of the non-convex energy landscape near the warm-start solution.

D. Linear Ramp Parameterization

• To avoid the computationally expensive and often unstable classical optimization of QAOA variational parameters ($\gamma, \beta$), the authors employ a deterministic linear ramp schedule.
• Parameters are initialized based on an adiabatic evolution concept: $\gamma_k$ increases linearly while $\beta_k$ decreases linearly across the $p$ layers.
• A small set of ramp rates ($\Delta$) is tested, and the best performing one is selected, eliminating the need for a classical outer optimization loop.

3. Key Contributions

1. Generalized HUBO Formulation: The paper presents a direct mapping of Gray-coded M-QAM MIMO detection to a quantum Hamiltonian without requiring oracle-based assumptions or QUBO reductions.
2. Hybrid WSLR-QAOA Framework: Introduction of a novel algorithm combining:
   • Warm-Start: Using BM-BCD for high-quality initialization.
   • Custom Mixer: A problem-specific mixer operator.
   • Linear Ramp: A parameter strategy that bypasses costly classical optimization.
3. Scalable Classical Solver: The use of BM-BCD for the SDP relaxation makes the warm-start generation feasible for large-scale MIMO systems where traditional solvers fail.
4. Hardware Validation: The framework is evaluated not only in simulation but also on real quantum hardware (IBM Fez processor, Heron r2 architecture), incorporating realistic noise models (depolarizing noise, readout errors).

4. Results

The paper provides extensive simulation and hardware results for 2x2 and 4x4 MIMO systems with 16-QAM:

• Symbol Error Rate (SER) Performance:
  • WSLR-QAOA achieves SER performance close to the optimal ML detector across all Signal-to-Noise Ratio (SNR) values.
  • It significantly outperforms classical linear detectors (ZF, MMSE) and standard QAOA variants.
  • In 4x4 MIMO scenarios, standard QAOA fails due to an under-parameterized ansatz, whereas WSLR-QAOA remains robust and scalable.
• Energy Landscape:
  • The combination of warm-start initialization and the problem-informed mixer results in a smoother cost landscape with lower mean cost and reduced variance compared to standard QAOA. This indicates a more stable optimization process.
• Hardware Robustness:
  • On the IBM Fez processor, the algorithm maintains competitive accuracy.
  • At low SNR, the hardware results are nearly identical to the ML bound.
  • At higher SNR, performance degrades moderately due to accumulated gate errors and decoherence but remains superior to other quantum baselines.
• Convergence: The linear ramp strategy allows the algorithm to converge faster than standard QAOA, which requires iterative parameter tuning.

5. Significance

This work represents a significant step toward the practical deployment of quantum algorithms in telecommunications:

• Bridging Theory and Practice: It moves beyond theoretical BPSK models to address the complex, non-convex reality of modern Gray-coded M-QAM systems.
• NISQ Viability: By demonstrating success on actual IBM hardware with noise, it proves that hybrid quantum-classical approaches can provide value even before fault-tolerant quantum computers are available.
• Scalability: The integration of efficient classical preprocessing (BM-BCD) with quantum optimization offers a pathway to handle large-scale MIMO detection problems that are currently intractable for purely classical ML detectors.

In summary, the paper establishes that a Warm-Start Linear-Ramp QAOA is a viable, robust, and high-performance solution for next-generation MIMO data detection, effectively leveraging current NISQ hardware capabilities.