**6. Simulation Results**

Since the traditional MUSIC algorithm cannot solve the multi-source tracking problem when target number is varying, this paper uses FLOM matrix to substitute the covariance matrix to obtain the corresponding MUSIC spatial spectrum, which can be as the particle likelihood function. We proposed a UT-MeMBer DOA tracking algorithm under RFS framework, which can be named as UT-MB-FLOM-MUSIC algorithm. The Generalized Signal to Noise Ratio (GSNR) is defined as

$$\text{GSNR} = 10 \log \left( \mathbb{E} \left( \left| \mathbf{s}(k) \right|^2 \right) / \gamma \right) \tag{29}$$

where γ represents the noise dispersion parameter, and GSNR represents the ratio of signal intensity and noise dispersion. In the simulation, different characteristic indices α describe the degree of impact of different noises.

In the following simulation experiments, the estimated performance is evaluated by the root mean square error (RMSE), which is defined as

$$\text{RMSE} = \frac{1}{P} \sum\_{p=1}^{P} \frac{1}{\text{MC}} \sum\_{j=1}^{\text{MC}} \left( \sqrt{\frac{1}{K} \sum\_{i=1}^{K} \left( \mathbf{x}\_{ij} - \overline{\mathbf{x}}\_{ij} \right)^2} \right) \tag{30}$$

where *xij* and *xij* represent the estimated values and real values of the azimuth angle in the *j*th Monte Carlo (MC) simulation experiment at time *i*, respectively, and *P* indicates the number of sources at time *i*. .

Assuming that the sources *<sup>x</sup><sup>k</sup>* <sup>=</sup> θ*k*(*t*), θ*k*(*t*) *T* move with a constant velocity . θ*k*(*t*) rad/s, the constant velocity (CV) model is given as

$$\mathbf{x}\_{k} = \mathbf{F}\_{k}\mathbf{x}\_{k-1} + \mathbf{G}\mathbf{v}\_{k} \tag{31}$$

where the transfer matrix *F<sup>k</sup>* and *G* are defined by

$$\mathbf{F}\_k = \begin{bmatrix} 1 & \Delta T \\ 0 & 1 \end{bmatrix}; \mathbf{G} = \begin{bmatrix} \Delta T^2 / 2 \\ & \Delta T \end{bmatrix} \tag{32}$$

respectively, where Δ*T* = <sup>1</sup>*s* denotes the time step, and *v<sup>k</sup>* is a zero-mean real Gaussian process used to model the disturbance on the source velocity, i.e., *<sup>v</sup><sup>k</sup>* ∼ N(**0**, **<sup>Σ</sup>***k*) with **<sup>Σ</sup>***<sup>k</sup>* <sup>=</sup> 1.

Experimental conditions are as follows: The number of array elements is *M* = 10, *d* = λ/2, the observation time is *K* = 50 s , *L* = 100, GSNR = 10 dB, MC = 100, and ξ = 5. The source survival probability *ps*,*k*(*xk*) = 0.99, and the source detection probability *pD*,*k*(*xk*) = 0.98. In the UT-MB-FLOM-MUSIC algorithm prediction step, we assume that there are six new sources at each time, i.e., *JB*,*<sup>k</sup>* = 6, all obeying a uniform distribution on [−π/2 , π/2] and each new source produces 300 particles, i.e., *NB*,*<sup>k</sup>* = 300. In the update step, the MUSIC spatial spectral function is used to replace the likelihood function and is exponentially weighted, which improves the feasibility of the algorithm. In the impulse noise model, the noise is Gaussian noise when α = 2. The DOA estimation method

based on the MeMBer can be named as MB-MUSIC algorithm, and the DOA estimation method based on the MeMBer of FLOM vector can be named as MB-FLOM-MUSIC algorithm.
