Performance Analysis of Decision Directed Maximum Likelihood MIMO Channel Tracking Algorithm

In this paper, the performance of decision directed (DD) maximum likelihood (ML) channel tracking algorithm is analyzed. The ML channel tracking algorithm presents efficient performance especially in the decision directed mode of the operation. In this paper, after introducing the method for analysis of DD algorithms, the performance of ML Multiple-Input Multiple-Output (MIMO) channel tracking algorithm in the DD mode of operation is analyzed. In this method channel tracking error is evaluated for given decision error rate. Then, the decision error rate is approximated for given channel tracking error. By solving these two derived equations jointly, both the decision error rate and the channel tracking error are computed. The presented analysis is compared with simulation results for different channel ranks, Doppler frequency shifts, and SNRs, and it is shown that the analysis is a good match for simulation results especially in high rank MIMO channels and high Doppler shifts.

💡 Research Summary

The paper presents a rigorous analytical framework for evaluating the performance of a decision‑directed (DD) maximum‑likelihood (ML) channel tracking algorithm in multiple‑input multiple‑output (MIMO) systems. Traditional channel estimation techniques for time‑varying wireless links either rely on frequent pilot symbols (non‑DD) or employ sub‑optimal adaptive filters, both of which suffer from high overhead or limited accuracy under fast fading. By contrast, the DD‑ML approach re‑uses previously detected data symbols to update the channel estimate, thereby reducing pilot overhead while retaining the optimality of an ML criterion.

The authors first describe the system model: the received vector at time k is (y_k = H_k s_k + n_k), where (H_k) is the unknown channel matrix, (s_k) the transmitted symbol vector, and (n_k) complex Gaussian noise. In DD mode the algorithm employs the previously detected symbol (\hat{s}_{k-1}) to form the ML estimate (\hat{H}k = \arg\min_H |y_k - H\hat{s}{k-1}|^2), which reduces to a linear least‑squares solution using the Moore‑Penrose pseudo‑inverse.

The core contribution lies in establishing two coupled analytical expressions that link the mean‑square channel tracking error (\sigma_H^2) and the symbol decision error probability (P_e). The first expression derives (\sigma_H^2) as a function of the Doppler‑induced temporal correlation (modeled by Jakes’ autocorrelation (R_H(\tau)=J_0(2\pi f_D\tau))), the noise variance, and an additional term that captures error propagation from incorrect decisions. The second expression approximates (P_e) by treating the effective signal‑to‑noise ratio after channel estimation as (\frac{E_s}{N_0+\sigma_H^2E_s}) and applying the standard Q‑function approximation for M‑ary constellations. Consequently, the pair of equations:

\